Quick update for those interested: I now have a prototype Miocene ACCESS-ESM1.5 simulation running for 1 month without error . Some things I did in this iteration:
- Instead of replicating a single cell of land to all new land cells (which worked for Drake Passage), I used nearest neighbour extrapolation for the new land cells. Some values looked a bit funny with single cell extrapolation, e.g. soil temperature in some new Arctic cells looked very warm compared to neighbouring values… I had a feeling this was causing instabilities.
- After that change, instead of crashing after 7 timesteps, the model completed ~20 timesteps. The key difference then was that the error messages were then coming from the ocean, not the atmosphere.
- I changed my ocean_temp_salt.res.nc restart file, which was previously copied from the pre-industrial (with extrapolation to new ocean cells). I greatly simplified it by changing temperature to 10 degC everywhere and salinity to 34.7 psu everywhere. This got past the ocean crash point, and now I have a 1-month successful simulation. So there might be some potential for new ocean instabilities in my new geography, but I’ll deal with that later!
- Earlier in the process, I also flattened the orography to 10 m everywhere to see if that would help (it made no difference - the model crashed just the same). I will now need to update the model to have proper orography…
- I haven’t touched the river runoff - it’s still on pre-industrial settings… so that needs to be changed.
- I’m also going to make some tweaks to my land-sea mask again before I go further, because I want to change some stuff I did in the first iteration.
I will provide code and example files for this stuff… just want to make some revisions first.
@Aidan @MartinDix @atteggiani @Dietmar_Dommenget @abhik @paulleopardi @edoddridge @LaurieM @Katrin @gpontes @HIMADRI_SAINI