#20 Restarting EC-Earth from an existing experiment with different domain decomposition

Hi @pbarriat and @ddocquie,

I'm trying to use the feature of EC-Earth to start a simulation from pre-existing restarts. This seems feasible as long as the domain decomposition from the mother run is the same as that of the child run.

In case the domain decompositions do not match, I can still make a work around for NEMO using rebuilt restarts from the mother run. But do you know about IFS?

If no one knows, I'll probably use a tweak and pretend starting from scratch while replacing the ocean initial files with restarts and keeping the default IFS initial files.

Hi @fmasson, sorry but I've never used a different domain decomposition when starting EC-Earth from a pre-existing restart.

Hi @fmasson,

You can't, "the restarted run must always have the same parallel decomposition (TASKS x THREADS) otherwise the restart will fail."

IFS restart files are written as unformatted binary: How to restart the OpenIFS model forecast

All the spinup runs I already used before came with initial conditions for the atmosphere which are nothing more than the official initial conditions (for 1990-01-01) with dates changed.

So you have a risk of a weak instability in the first months of the simulations...

OK, thanks!

I think I'll setup my work around then: pretend starting from scratch but with "fake" 3-D ocean fields obtained from restarts.

I'll keep you updated.

That solution is finally too intrusive, it requires to not use the interpolation weights as this done when true initial data is used. I'll rather use the same domain decomposition as @ddcoquie EC05 and start from there.

I found it even very difficult to restart an experiment from someone else's restarts even if the domain decompositions match.

I tried:

• Starting from David's IFS and NEMO restarts --> failed
• Starting from my NEMO restarts but from David's IFS restarts --> failed
• Starting from my IFS restarts but from David's NEMO restarts --> suceeded.

So if an experiment is to be started from results of another one, the best would be

(1) to run the model from scratch for one year (2) to run the model from the IFS restarts that were generated, but overriding the NEMO restarts using the ones from the mother experiment.

Hi @fmasson, this is strange. I've never had problems in restarting from an existing experiment. But if I remember correctly, I always keep the number of processors similar to the 'mother experiment'. You can have a look at my .sh files on zenobe: /home/acad/ucl-elic/ddocquie/modeles/ecearth/runtime/classic. For example, you can compare EC05.sh (150-yr long control run that you're using) and AL00.sh (1-yr long albedo perturbed experiment starting at year 108 of EC05 with reduced albedo, this is one of the members used by Steve in his master thesis). I'll be in LLN tomorrow, we can discuss it.

Hi @ddocquie,

Perhaps I should have been more precise. I can indeed restart from another run of mine, but I can't from your EC05 run. The problem is maybe that our two versions are not identical (I'm using 3.3.0).

François

Yes, that may be the reason. Worth checking with @pbarriat

OK, I'm going to check that!

Technically, the release should not impact a restart.

The goal is to restart a run from the EC05 @ddocquie run, right @fmasson ?

Yes, restarting an EC-Earth 3.3.0 experiment from the last restart of David's EC05 would be useful.