TM matrix for Mercator's GLORYS dataset ?

[simonweppe]

Hi there and thanks for the great work.

I have limited experience with the TMM method so please excuse if irrelevant :)

Was there any attempts to build TM matrix from Mercator's GLORYS output (for example from last 10-year mean) ?
It could significantly increase the spatial resolution of predicted outcomes (at least for "current" climatic conditions). I was not able to find any application of this so I wonder if I might missing something ?

Provided we are able to source to correct variables for water mass transport (saw some post of interpolation challenges), in theory we could build such a TM with your code, if I understand well ?

Best,
Simon

[briochemc]

Hi Simon,

I'm not aware of any attempt to build a transport matrix from GLORYS. But yes, in theory, it's possible.

However, note that transport matrices typically "shine" at lower resolutions (I have not dealt with anything much finer than 1° myself). Unless I'm mistaken, with its 8km (1/12°) resolution and 50 levels, we are talking of a matrix size of about 300,000,000 for GLORYS, which might be a problem.

I'd be happy to help out but I'm currently looking for a job and uncertain I will be able to allocate much time to this when I find one! Hopefully one day we can get TMIP funded and build the tooling to easily extract transport matrices from every ocean model out there!

Hope this helps,
Cheers,
Benoit

PS: The "Transport Matrix Method (TMM)" usually refers to Khatiwala's method for building matrices (Khatiwala, 2005, 2007), which is more accurate but much more complicated and expensive than the method employed here in this package:

• The TMM modifies the online model code to inject tracers at many places simultaneously and record their transport as impulse response functions to build the matrix. Hence you must (i) know your way around the code to add the required hooks, and (ii) run the online model. (side note: I know of at least one TMM work for a high-resolution model (FEOTS; Schoonover et al., 2023), although this may not be relevant for you.)
• The method used here is simpler: Build the matrix directly from the archived output of velocities. Although there still is some work to calibrate the diffusion part, this method is much simpler than the TMM. And much cheaper too as you don't need to run the online model.

[simonweppe]

Thanks @briochemc - I appreciate the infos and insights.

Indeed, I just realized as I dived deeper in the literature that your approach was somewhat different than Khatiwala's TMM...We'd be more inclined to the approach used in your package as it's much lighter/quicker to implement.

Will keep you posted if we make good progress on this. Getting the mean fields would be no problem...let's see how things go with a 300,000,000 matrices though !