Tips on chunking multi-year runs into individual years?

[Veronica Tamsitt]

Hi all,

I am trying to break down longer CMS runs into individual years, then concatenate the trajectory files after the fact in case of crashing and to avoid running into computing limitations (I haven't been able to get the restart option to work in the CMS code).

I saw in a previous thread that others use this method. To see that this is working correctly I did a test where I did a 2-year continuous run, then ran the 1st and 2nd year as separate runs, releasing the particles in the second run at the locations from the last output timestep of the previous year and joining together. 

Unfortunately the trajectories do not match up exactly (although many are very similar), I've attached a couple of plots of lon/lat/depth of particles in the continuous and individual year cases for the time steps just before and after the transition (timestep 73 is the concatenation point).

There are two possibilities I can think of for why they don't match:

1) I have not timed the second release year correctly to match up with the first year. Currently I am releasing on Jan 3, 1980, 0 seconds (and at 5 day intervals thereafter), running for 1 year with a 90 minute timestep and 5-day output, so the last output position in the traj file is exactly 1 year since release, Jan 3, 1981 (I believe). So I release the next year-long run on Jan 3, 1981 using the previous end position.

2) the two experiments will not produce identical output because of some numerics in CMS when restarting. But as some of the particle trajectories drift slightly apart, and others are much further off, I'm doubtful this is the case.

If anyone has experience with this and knows how to get it working right it would be much appreciated.

Thanks,

Veronica