COSIMA / mom6-panan

Pan-Antarctic regional configuration of MOM6
MIT License
6 stars 6 forks source link

Drift in spin up #15

Closed schmidt-christina closed 1 year ago

schmidt-christina commented 2 years ago

I analysed the abyssal overturning, Drake Passage transport and zonal mean T/S at 4500 m at several latitudes in the panantarctic 1/10° z* configuration and compared it to the global ACCESS-OM2-01 RYF. The drift in panan is much stronger (probably 3x for the abyssal overturning and Drake Passage transport) compared to the global ACCESS-OM2-01 RYF run, even close to the northern boundary, and far from spun up after 40 years.

image

image (1)

image (2)

AndyHoggANU commented 2 years ago

Yes, I see the issue here. I wonder whether you have compared this with the 10-year output from the original pan-ant at 0.05°? I know these are short, but you should still see the early signs of such big drifts...

schmidt-christina commented 2 years ago

Here are plots with the original pan-ant at 0.05° but that doesn't make things better, even a bit more complicated because trends can be in the opposite direction.

image

image

schmidt-christina commented 2 years ago

Next Steps:

adele-morrison commented 2 years ago

@AndyHoggANU do you have any free time between 9-11 tomorrow to join the panan hackathon? One of our main priorities is to get the global 1/10deg MOM6 run going, which currently has your name against it. If you can't join, perhaps you could give us a quick update / pointer to what you've thought about / prepared so far?

angus-g commented 2 years ago

That one might be partially waiting on me to generate the IC, I was going to get to that today, so we’ll definitely have an idea of how much extra work the global will require by tomorrow morning!

adele-morrison commented 2 years ago

Awesome, thanks @angus-g!

schmidt-christina commented 2 years ago

Here is the path to the data from the panan-01 with smoothed initial conditions that I ran for 2 years:

/scratch/e14/cs6673/mom6/archive/panan_01deg_jra55_ryf_smooth_init

I haven't looked at it yet and will do that after CLEX.

AndyHoggANU commented 2 years ago

OK kids — seems @angus-g has fixed our problem with the global runs. It turns out that our v-points are exactly on the Equator in the ACCESS-OM2 grid, and there was a divide-by-f operation somewhere which went to NaNs. Now fixed and have run the first three months of a global 0.1°. Hopefully will be a full year by the morning and I’ll stash it somewhere so that you can all see the output.

AndyHoggANU commented 2 years ago

First year is done and is up here: /g/data/ik11/outputs/mom6-global-01/global-01-v1 I haven't looked at anything, but the diagnostics table should be similar to panant -- so if anyone manages to run some scripts please let us know.

AndyHoggANU commented 2 years ago

I managed to get through 4 years of the global simulation, but I think I overdid the tilmestep, so have pulled it back to year 2 and restarted. There should be something interesting there by the end of the week ...

AndyHoggANU commented 2 years ago

Global run is now up to year 6 and is looking more sensible now that I dropped the tilmestep back down. Note that I still get V_velocity truncations in a strait just south of Yuzhny Island which I'm prepared to ignore for now.

I will keep this going until at least year 10, but if you want to have a quick look at the output, see here: https://github.com/AndyHoggANU/mom6-diagnostics/blob/main/global-01/ComparisonWithPanant.ipynb

Note that on 4416 cores this simulation takes about 13.5 wall time hours per model year (slightly slower than ACCESS-OM2 on ~5000 cores) and takes about 120 kSU/year (about the same as ACCESS-OM2), noting I am using a slightly longer tilmestep and we are yet to do any optimisation on MOM6-SIS2 ... So, not bad for a first attempt.

aekiss commented 2 years ago

Nice to hear it's performing well.

We put Rayleigh damping in the Kara Strait south of Yuzhny Island to deal with it blowing up there in ACCESS-OM2-01 https://github.com/COSIMA/01deg_jra55_iaf/blob/master/ocean/field_table#L75-L122

schmidt-christina commented 1 year ago

I have included the global MOM6 at 1/10 now in the plots below. The abyssal overturning looks stable, but the drift in panan-01 also occurred mostly afterwards. The Drake Passage transport has a fairly similar drift as panan-01 within the first 10 years.

Based on this it seems that the boundary in panan-01 is not the cause of this drift but maybe different parameters in MOM5 vs MOM6?

image

image

schmidt-christina commented 1 year ago

Here are now the updated plots with 68 years for panan-01 and 30 years for mom6-global-01:

Overall, panan-01 and mom6-global-01 behave very similar (i.e. they drift) and mom6-global-01 is far away from equilibrium.

The divergence between panan-01 and mom6-global-01 in the abyssal overturning at 65°S after about 15 years was only temporarily and now mom6-global-01 shows the same (or even stronger) drift as panan-01.

The drift in Drake Passage transport continues with a similar rate.

Zonally averaged T and S at 4500 m show the same evolution in panan-01 and mom6-global-01, expect at 37°S due to the boundary there in panan-01.

image

image

image

adele-morrison commented 1 year ago

Thanks @schmidt-christina that's useful to know. So are others happy to conclude from this that the drift in panan is not due to the boundary and is just due to the different equilibrium states of ACCESS-OM2-01 and MOM6-global-01?

If yes, then probably we don't need to run the MOM6-global-01 with ACCESS-OM2-01 initial conditions any further out? We could shift our efforts and computer time towards the MOM6-global-01 spinup with WOA initial conditions?

AndyHoggANU commented 1 year ago

Oh, yeah, I agree. Thanks @schmidt-christina - I will kill the current run and restart from WOA. Will keep you posted.

PaulSpence commented 1 year ago

Re: "We could shift our efforts and computer time towards the MOM6-global-01 spinup with WOA initial conditions?"

Yep, I agree as well. Thanks Christina!

P

On Wed, Jan 11, 2023 at 8:45 AM Andy Hogg @.***> wrote:

Oh, yeah, I agree. Thanks @schmidt-christina https://github.com/schmidt-christina - I will kill the current run and restart from WOA. Will keep you posted.

— Reply to this email directly, view it on GitHub https://github.com/COSIMA/mom6-panan/issues/15#issuecomment-1377928966, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABSWJXDTFP5JNHEE22PBAXLWRXJ7XANCNFSM6AAAAAAQYNXMVE . You are receiving this because you are subscribed to this thread.Message ID: @.***>

-- I’m sending this message at a time that suits me. I don’t expect you to reply outside of your own work hours.

Paul Spence, Assoc. Prof. ARC Future Fellow Institute for Marine and Antarctic Studies https://www.imas.utas.edu.au/ University of Tasmania, Hobart, Australia https://paulspence.github.io/

AndyHoggANU commented 1 year ago

Hey @angus-g -- to do this I need to pick up the original ACCESS-OM2 01deg initial Temp/Salt file ... currently I'm running using the /g/data/x77/ahg157/inputs/mom6/global-01/ inputs, which I don't have permission to copy a restart file to.

Can you copy the relevant restart file in there for me? I guess it is this one: /g/data/ik11/inputs/access-om2/input_20201102/common_01deg_jra55

angus-g commented 1 year ago

I've put the ocean_temp_salt.res.nc in there as ocean_temp_salt.access.nc, hopefully that's the right one?

AndyHoggANU commented 1 year ago

Hopefully. Will give it a burl.

AndyHoggANU commented 1 year ago

Seems to be running, but I won't have time to check on it today -- @schmidt-christina , if you want to have a look as it's going it will be synced with global-01-v2. Should have some output there by lunchtime today.

AndyHoggANU commented 1 year ago

Just to note - I've had a quick look at a few global metrics in this -v2 simulation, and looks OK so far. Here is DP transport:

Screenshot 2023-01-17 at 3 00 46 pm
adele-morrison commented 1 year ago

I'm confused - the panan also doesn't seem to have a trend over this period. Is it just a shorter time span compared to the plot above with the large trend? Does that mean we probably need longer to see if there's a trend in the global-01-v2 also?

AndyHoggANU commented 1 year ago

True. Yes. Yes.

Screenshot 2023-01-17 at 3 47 41 pm
willaguiar commented 1 year ago

Getting in the middle of the discussion here, but maybe this info might help ( or maybe you all already know...).

The OM4.0 (MOM6 + SIS, 1/4 ocean) when ran in z have a pretty big drift with a mid-depth warming trend below 500m (Figs 8c red line and 9d, of Adcroft et al 2019). This drift was fixed by running MOM6 with hybrid isopycnal/z coordinates. I don't know if global-01-v2 is being ran with z*, but if it does perhaps the same is happening here?

PS: I believe the authors blame the mid-depth warming on an inefficient diapycnal mixing in MOM6 under z*.

AndyHoggANU commented 1 year ago

OK, so it could be worth looking at the vertical distribution of heat as the model spins up. Will try to get to that but if anyone beats me there, please go for it.

Having said that, in ACCESS-OM2 the drift in heat content was opposite for 0.25° and (0.1° or 1°), so not sure we will see the same trend as in OM4 ...

adele-morrison commented 1 year ago

The drift in the mom6-global-01 is not looking great:

Screen Shot 2023-01-31 at 8 46 01 pm

The Southern Ocean abyss is warming and salinifying (blue line), with a similar drift to the panan-01-zstar (green line; likely not enough bottom water production). This is the opposite drift to ACCESS-OM2-01-RYF (orange), which cools and freshens in the Southern Ocean abyss (likely too much bottom water production). The panan drift is also sensitive to vertical coordinate and resolution, with the 01-hycom (red) and 005-zstar (not shown here) having opposite drifts compared with the panan-01-zstar...

In summary, I think it is going to be hard to find a decent model simulation that is in equilibrium to use to force the new panan at the boundary. Any model we use is going to be biased compared to obs. And the equilibrium case we might use for forcing at the boundary (e.g. mom6-global-01) will not necessary have a similar equilibrium state to that of the 1/20th panan or a panan with different vertical coords.

However, Andy thought of a good idea this morning to get around this problem. We can use a year of output from very early in the spinup (like year 2) of either ACCESS-OM2-01 or OM4 to force at the northern boundary. That way we have consistent velocity and stratification data for the boundary that is relatively close to equilibrium. Another benefit of this is that we don't have to wait for the mom6-global-01 to spinup.

What do others think? Any issues with this?

PaulSpence commented 1 year ago

Re: We can use a year of output from very early in the spinup (like year 2) of either ACCESS-OM2-01 or OM4 to force at the northern boundary. That way we have consistent velocity and stratification data for the boundary that is relatively close to equilibrium. Another benefit of this is that we don't have to wait for the mom6-global-01 to spinup.

So the idea here is that the T/S will be close to obs and the velocities will be just spun up enough to be reasonable? I guess the dangerous assumption is that the boundary vels are reasonably spun up by year 2. Not the worst assumption, but does Fig 4 from Kiss et al look a little scary, e.g. looney DP transport for >20yrs after initialization:

Screen Shot 2023-02-01 at 12 06 21 pm

?

adele-morrison commented 1 year ago

Yes velocities at the boundary should match the initial T/S structure, so we should get a reasonable overturning. But we don’t want to wait so long that the Drake Passage transport (and hence abyssal stratification) has a large bias like in the above figure.

On Wed, Feb 1, 2023 at 2:09 PM, Paul Spence @.***> wrote:

Re: We can use a year of output from very early in the spinup (like year 2) of either ACCESS-OM2-01 or OM4 to force at the northern boundary. That way we have consistent velocity and stratification data for the boundary that is relatively close to equilibrium. Another benefit of this is that we don't have to wait for the mom6-global-01 to spinup.

So the idea here is that the T/S will be close to obs and the velocities will be just spun up enough to be reasonable? I guess the dangerous assumption is that the boundary vels are reasonably spun up by year 2. Not the worst assumption, but does Fig 4 from Kiss et al look a little scary: [image: Screen Shot 2023-02-01 at 12 06 21 pm] https://user-images.githubusercontent.com/6644956/215935741-f3fc37fb-cbfa-4c9e-aa28-63afb1095222.png ?

— Reply to this email directly, view it on GitHub https://github.com/COSIMA/mom6-panan/issues/15#issuecomment-1411398241, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACA44UZUJAQ2DMBMYRVA233WVHHVTANCNFSM6AAAAAAQYNXMVE . You are receiving this because you commented.Message ID: @.***>

adele-morrison commented 1 year ago

Actually, I just had a little more of a look at the abyssal overturning at 37S in OM4 compared with ACCESS-OM2-01, and am doubting if ACCESS-OM2-01 is still the best choice for boundary forcing.

Here's the plot we looked at this morning, where they look pretty similar:

Screen Shot 2023-02-03 at 11 22 31 am

But if you look at annual averages is shows up much more clearly than the abyssal overturning in ACCESS-OM2-01 is really weak even right at the start:

Screen Shot 2023-02-03 at 11 24 25 am

I don't understand why there's such a huge immediate decline in ACCESS-OM2-01. Anyone have any insight? And opinions if we should use OM4 instead?

willaguiar commented 1 year ago

Forcing with OM4 could be a good option because is tested and documented already. The warming trend I mentioned before in OM4 is about ~0.005C/yr, so I think it wouldn't be an issue if using the first few years of a global simulation.

adele-morrison commented 1 year ago

Closing this with a note that we decided to force at the boundary with year 2 of OM4.