Oceananigans.jl: Support for a full equation of state for seawater (TEOS-10?)

[ali-ramadhan]

@johncmarshall54 has suggested we switch to a full equation of state for seawater for LESbrary simulations as temperature and salinity profiles from state estimates may not be statically stable in Oceananigans when using a linear equation of state. The Roquet et al. (2015a) idealized nonlinear equation of state should be better. Ideally we would use the same equation of state as ECCO/SOSE I think, but TEOS-10 should be what we aim for I suppose.

I think we have a couple of options:

1. There's a Julia wrapper for the TEOS-10 C library but this probably won't work on the GPU: https://github.com/ax1ine/GSW.jl
2. A 6+52-term polynomial approximation to TEOS-10, accurate to ~0.1% and suitable for Boussinesq models, from Roquet et al. (2015b). We can code up a pure Julia version as there already exist Python, MATLAB, and Fortran implementations here: https://github.com/fabien-roquet/polyTEOS (Thank you @fabien-roquet!)

In implementing a pure Julia TEOS-10 equation of state we should make sure it can be shared between Oceananigans and CliMA Ocean (cc @blallen).

cc @jm-c @christophernhill @glwagner @rafferrari who might have suggestions or comments.

@leios Do you think evaluating a 52-term polynomial at every grid point will be problematic on the GPU? I imagine GPUs should be fast at evaluating polynomials but maybe we have to be smart about storing the coefficients in memory?

References:

1. Roquet et al. (2015a), "Defining a Simplified Yet “Realistic” Equation of State for Seawater", DOI: https://doi.org/10.1175/JPO-D-15-0080.1
2. Roquet et al. (2015b), "Accurate polynomial expressions for the density and specific volume of seawater using the TEOS-10 standard", DOI: https://doi.org/10.1016/j.ocemod.2015.04.002

[glwagner]

I also think it’s a good idea to use a full equation of state for all simulations sooner rather than later. It’s simpler: we won’t have to report constants of linearization everywhere. And setting up simulations will be easier.

I’ll defer to the modelers for whether Roquet’s approximation is an acceptable model for TEOS-10.

This package is relevant and we should consider contributing to it rather than implementing an equation of state somewhere in the Clima ecosystem:

https://github.com/gher-ulg/PhysOcean.jl

[blallen]

In implementing a pure Julia TEOS-10 equation of state we should make sure it can be shared between Oceananigans and CliMA Ocean (cc @blallen).

:100: :100: :100: :100: :100: :100: :100: :100: :100: :100: :100: :100: :100:

This package is relevant and we should consider contributing to it rather than implementing an equation of state somewhere in the Clima ecosystem:

https://github.com/gher-ulg/PhysOcean.jl

:sunglasses: :sunglasses: :sunglasses: :sunglasses: :sunglasses: :sunglasses: :sunglasses: :sunglasses: :sunglasses: :sunglasses: :sunglasses: :sunglasses: :sunglasses:

Whatever we do should be shared between Oceananigans and Clima. Either a custom implementation or using the PhysOcean.jl package is fine by me.

[rafferrari]

The Roquet’s approximation is perfectly sufficient for Oceananigans, because it will never be used for global calculations where local approximations are an issue. However I agree with everybody else that it would be best to use the same EOS in Ocenanigans and Climate_Ocean. in that case we should adopt TEOS-10. Be warned that it is quite inefficient through. So we may be hit performance-wise. Hard to tell without trying.

Raffaele

On Mar 9, 2020, at 11:48 AM, Gregory L. Wagner [e-mail] wrote:

I also think it’s a good idea to use a full equation of state for all simulations sooner rather than later. It’s simpler: we won’t have to report constants of linearization everywhere. And setting up simulations will be easier.

I’ll defer to the modelers for whether Roquet’s approximation is an acceptable model for TEOS-10.

This package is relevant and we should consider contributing to it rather than implementing an equation of state somewhere in the Clima ecosystem:

https://github.com/gher-ulg/PhysOcean.jl https://github.com/gher-ulg/PhysOcean.jl
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub https://github.com/climate-machine/Oceananigans.jl/issues/692?email_source=notifications&email_token=AK24ROIBMZMYD77AGEZNM3DRGUMURA5CNFSM4LEKJAAKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEOHZCYQ#issuecomment-596611426, or unsubscribe https://github.com/notifications/unsubscribe-auth/AK24ROONT4RO4YCCF6BTIIDRGUMURANCNFSM4LEKJAAA.

[johncmarshall54]

What does J-M think?
John

On Mon, Mar 9, 2020, 3:07 PM rafferrari [e-mail] wrote:

The Roquet’s approximation is perfectly sufficient for Oceananigans,
because it will never be used for global calculations where local
approximations are an issue. However I agree with everybody else that it
would be best to use the same EOS in Ocenanigans and Climate_Ocean. in that
case we should adopt TEOS-10. Be warned that it is quite inefficient
through. So we may be hit performance-wise. Hard to tell without trying.

Raffaele

On Mar 9, 2020, at 11:48 AM, Gregory L. Wagner [e-mail]
wrote:

I also think it’s a good idea to use a full equation of state for all
simulations sooner rather than later. It’s simpler: we won’t have to report
constants of linearization everywhere. And setting up simulations will be
easier.

I’ll defer to the modelers for whether Roquet’s approximation is an
acceptable model for TEOS-10.

This package is relevant and we should consider contributing to it
rather than implementing an equation of state somewhere in the Clima
ecosystem:

https://github.com/gher-ulg/PhysOcean.jl <
https://github.com/gher-ulg/PhysOcean.jl>
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub <
https://github.com/climate-machine/Oceananigans.jl/issues/692?email_source=notifications&email_token=AK24ROIBMZMYD77AGEZNM3DRGUMURA5CNFSM4LEKJAAKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEOHZCYQ#issuecomment-596611426>,
or unsubscribe <
https://github.com/notifications/unsubscribe-auth/AK24ROONT4RO4YCCF6BTIIDRGUMURANCNFSM4LEKJAAA
.

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/climate-machine/Oceananigans.jl/issues/692?email_source=notifications&email_token=AKXUEQUJLIMSIC2HXXYDQG3RGU47RA5CNFSM4LEKJAAKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEOIVI3Y#issuecomment-596726895,
or unsubscribe
https://github.com/notifications/unsubscribe-auth/AKXUEQV7RCQMEU6MW3UZPZTRGU47RANCNFSM4LEKJAAA
.

[ali-ramadhan]

What does J-M think?

Talked to @jm-c earlier today who agreed that the Roquet et al. (2015b) 52-term polynomial approximation to TEOS-10 is a good choice.

Although he did point out that ECCO and SOSE are probably using an older pre-TEOS-10 equation of state (from 1995 or 2003 based on the EOS-80 but for potential temperature I think?).

[ali-ramadhan]

The Roquet’s approximation is perfectly sufficient for Oceananigans, because it will never be used for global calculations where local approximations are an issue. However I agree with everybody else that it would be best to use the same EOS in Ocenanigans and Climate_Ocean. in that case we should adopt TEOS-10. Be warned that it is quite inefficient through. So we may be hit performance-wise. Hard to tell without trying.

Thanks for the feedback @rafferrari. I talked to @leios earlier today and we think it shouldn't be a problem on the GPU. It's just straight up number crunching so it might benefit from being run on a GPU. But we can make sure by doing a quick benchmark.

[jm-c]

Ali,

This is a correct transcript of what I said earlier.

Cheers,
Jean-Michel

On Mon, Mar 09, 2020 at 07:21:21PM -0700, Ali Ramadhan wrote:

What does J-M think?

Talked to @jm-c earlier today who agreed that the Roquet et al. (2015b) 52-term polynomial approximation to TEOS-10 is a good choice.

Although he did point out that ECCO and SOSE are probably using an older pre-TEOS-10 equation of state (from 1995 or 2003 based on the EOS-80 but for potential temperature I think?).

--
You are receiving this because you were mentioned.
Reply to this email directly or view it on GitHub:
https://github.com/climate-machine/Oceananigans.jl/issues/692#issuecomment-596867518

[glwagner]

This is resolved as of v0.30.0!