Ocean spin-up in coupled simulations

[PeterHjortLauritzen]

Almost all our coupled simulations show global SST cooling in the first ~50 years (and associated redistribution of heat in ocean) regardless of the sign of RESTOM and then (more of less) stabilizes. Example below from CESM projects meeting slide:

Below is "slope chart" (regression) from @brianpm:

If all runs are going through a similar spin-up process, can we start our development runs from a ~50 year spun-up coupled simulation?

@klindsay28 suggested we do twin runs:

1. a standard initial run from the ATLAS initialization
2. a hybrid run starting from 54 at year 50 (*)

and see if we reach the same conclusion in both cases.

[PeterHjortLauritzen]

@adamrher @JulioTBacmeister @cecilehannay @gustavo-marques

I suggest we do the "twin" experiment with #75 (our latest "tuned" configuration) that is currently running. We could also choose a configuration that has already run 50+ years ...

[gustavo-marques]

@PeterHjortLauritzen, I just talked to @klindsay28 and we want to clarify what needs to be done for the "twin runs". Given that we will experiment with #75, we already have the two "a standard initial run from the ATLAS initialization": #75 and #73
We now need to do #73b and #75b, which will be hybrid runs starting from 54 at year 50.
Does that make sense?

[adamrher]

That makes sense to me. If the #75b run branched off from year 50 of #54, freezes immediately, at least we will have saved time in coming to that conclusion.

But have we not yet resolved the ocean initialization issue in hybrid runs yet #459?

[adamrher]

@brianpm some other runs that you might want to consider adding to your plot are the hybrid coordinate runs (47 and 55) and runs 26g and 46:

  dir =  (/"/glade/cheyenne/scratch/gmarques/archive/", \
           "/glade/cheyenne/scratch/gmarques/archive/", \
           "/glade/cheyenne/scratch/gmarques/archive/", \
           "/glade/derecho/scratch/hannay/archive/", \
           "/glade/scratch/gmarques/archive/", \

  fname = (/"b.cesm3_cam058_mom_e.B1850WscMOM.ne30_L58_t061.camdev_cice5.026g", \
            "b.e23_alpha16b.BLT1850.ne30_t232.046", \
            "b.e23_alpha16b.BLT1850.ne30_t232.047", \
            "b.e23_alpha16b.BLT1850.ne30_t232.054", \
            "b.e23_alpha16g.BLT1850.ne30_t232.055", \

In particular the hybrid coordinate runs warm in the first 25 years. I also wonder whether it would be instructive to do a scatter plot of climo RESTOM vs. the SST trend.

[adamrher]

This is pretty much the limit of my "colored time-series" approach for representing many coupled simulations (11 runs):

[brianpm]

I can easily add those other runs. I'll try to do that tomorrow and send an updated plot.

[brianpm]

Here is the updated plot with whichever of those runs have >25 years. In terms of the high-level finding that there's a cooling trend, nothing changes.

[swrneale]

I did want to mention that ENSO looks very different in the warm run (way too long period) and the original cold run (amplitude increasing again). But in the retuned 75 the initial look says it is back to something similar to 64. I guess some sensitivity to the significant SST changes would be expected.

[swrneale]

Also @brianpm where have you seen SST decreasing in all these runs? Is it systematic everywhere or is it regional, modal (i.e. PDO) looking?

[brianpm]

I didn't look at this, maybe others have. I just made the slope chart using global SST (removing land and ice-covered points).

[gustavo-marques]

@adamrher; yes, there is a short-term fix for ocean initialization, and I just posted instructions in #459

I am happy to run #73b and #75b. But before doing so, I have a question - should we do RUN_TYPE = startup instead of hybrid and initialize both the ocean and sea ice from run 54 at year 50? This way, it will be easier to compare against the corresponding twin by starting at the same time (0001-01-01). However, I am not completely sure about the purpose of doing hybrid runs instead of startup runs with initial conditions from restart files. Let me know your thoughts.

[adamrher]

My understanding is that both hybrid and startup runs reset the clock to 0001-01-01. Only branch runs start the clock according to the time on the restart files. With hybrid, it's easier to make the cice inic consistent since you just need to stage the rpointer.cice file pointing to the year 50 restart in the new run directory, and it will prioritize that over any default cice inic. Same for clm, etc ... so in general, hybrid runs will only select the default inic if there is no rpointer file staged in the run directory for that component. Does that clarify things?

[cecilehannay]

You could start an hybrid run with any start date you want: 0001-01-01, 0100-01-01, or what ever you want.

In this case, you should start with a start date of 0001-01-01.

[gustavo-marques]

Yes, thanks for clarifying. I was confused because RUN_STARTDATE: 0043-01-01in 064b, but I will keep the default 0001-01-01 in 73b and 75b.

[gustavo-marques]

Merci, @cecilehannay!
Runs 073b and 075b have started.

[adamrher]

Here's a first look at the new runs 73b and 75b:

[gustavo-marques]

@klindsay28 found an issue (or unintended experiment) in #73b and #75b. In both runs, I unintentionally kept the original initial conditions in the user_nl files for ice and land. So, only the ocean and atm are reading the restart from #054 at 0055-01-01. To inform us how the initial conditions of the sea ice affect the initial SST cooling, I will run 73c and 75c with all components reading the restart from 054.

[gustavo-marques]

@adamrher, can you please update your plot? I am curious to see how the Global SST cooling evolved in 75b.

[adamrher]

updated the plot above.

[cecilehannay]

We compared diags for:

• 75 vs 66 (yrs 50-70): diags
• 75c vs 66c (yrs 10-30): diags

Comparison is complicated by the fact that 66 freezes over.