We are running a nested model (using ERA5 data for the driving model) for 3 years - for which we are outputting a limited number of variables.
We want to be able to archive restart files at monthly intervals through these 3 years, so that we can come back and rerun any months we are particularly interested in, to output more variables if needed.
Can anyone advise on which files we need for these restarts (I’ve setup the archiving of the umnsaa_da024 file, which is written into the ics directory for each cycle - is this the file I need to archive?), and also how I can start a new model run from one of these restarts?
The suite I’m working in is u-cv012.
I’m not sure how to restart a nesting suite (maybe you have already figured it out - I see a
RESTART_DUMP in u-cv102)
Typically, a suite would just need a start file, lbcs, ancillary files, but presumably u-cv102 will need driving model data too.
I think we might have got this sorted. Because we’re using ERA5 driving data we have been able to ignore saving the driving model data - this just gets regenerated whenever we need it. So all we’ve saved is the restart dump at the end of each CRUN period. I copied the code for the N cycles that use this dump file for the next period, to create an option for the 1st cycle to read in a user defined dump file at the start of the model run.
Our tests have shown some small differences between the original and restarted model runs - mostly where there’s convective activity (our test suite is a 25km resolution setup). We’re trying out setting random number seed to carry through the restart dump too (option 2?) - hopefully this should give us a fully reproducibility.