How do we define the ASC?

[adele-morrison]

Issue: The location (i.e. which isobath) of the ASC varies spatially and temporally.

How do we deal with this?

Also, which quantity do we expect to be best correlated with heat transport? i.e. dynamically what aspect of the ASC matters most? Max speed? Structure? Transport?

Possibilities:

• Use the along-slope transport integrated between 2 isobaths.
• Max speed between 2 isobaths.
• Some measure of the structure, i.e. how surface intensified or otherwise is the ASC?
• Unsupervised classification methods.
• ....

Maybe we just try a few of these and see which works best? Then try to understand why?

[taimoorsohail]

I've done a simple GMM analysis on surface speed, \sqrt{u^2+v^2}, averaged over the last 10 years of the RYF simulation. Interestingly, for 2 clusters the model successfully highlights the extent of the ASC hugging the coast, though we need to do more analysis on whether this truly is the ASC or some other features that are being lumped together.

[wghuneke]

Interesting! It looks like a mixture of ASC and coastal current that is picked up by the 2 cluster version. And you need 5 or 6 clusters to include the Weddell Sea where surface velocities are low.

A little off topic: but interesting that the Weddell Gyre is a consistent feature for all higher clusters, while the Ross Gyre isn't.

[taimoorsohail]

That's a good point @wghuneke! So far I have just used the speed, so it doesn't take into account any directionality (which would clump things into different clusters). Perhaps n=4 is better because it removes the coastal current, and also highlights a weak current in the Ross and Weddell Seas... I'm guessing the Ross Gyre has more of a sub-surface signal, which is why it isn'y highlighted at the higher cluster numbers...?

One option for further exploration could be to use the full-depth speed profiles, or to add the surface temperature and salinity in our clustering algorithm. Or perhaps we just use the cluster outputs above as our definition and do diagnostics based on that?

[willaguiar]

Maybe another test could be to run the GMM analysis with along-slope speed instead of the actual speed, just for purposes of identification.

[taimoorsohail]

I'm not sure what you mean @willaguiar. The along-slope speed would be at a fixed contour (e.g., 1000m), so isn't that the current definition of the ASC anyway? I was thinking the purpose of using GMM was to objectively define the ASC without needing ad-hoc solutions like that? Or perhaps I don't understand the purpose of calculating the along-slope velocity!

Also, I did a similar clustering analysis by feeding in surface speed, temperature and salinity. Due to computational complexity I was only able to get 10 clusters. But basically the algorithm is more inclined to highlight surface water masses than speed, probably because 2 out of the 3 variables are T and S so they are outweighing the influence of speed in defining clusters.

[willaguiar]

You have a point @taimoorsohail . I thought of using the same velocity transposing process in the whole domain ( just extending northwards and southwards of the 1000m isobath) to make all U,V follow along/across coordinates instead of I,j in the whole domain. But I now realize it it would be tricky in the Antarctic peninsula.

Do you think "cleaning up" the domain before running the GMM could refine the results further? ( for example, only using the surface where the maximum ocean depth is smaller than 2500m , so the algorithm is forced to focus on the region of interest)

[taimoorsohail]

@willaguiar - yes - cleaning up would definitely help by masking the deeper regions. I had another idea - what if we performed the clustering of speed in isobath vs distance along contour space? That is, we calculate the speed along each isobath from 0 to 3000 or 4000m, and then cluster that speed to identify the isobaths along which the ASC signal is evident? This would have the advantage of identifying the isobaths over which averaging/integration should occur, so it would enable further analysis that isn't constrained to the 1000m isobath.

In Fig 5b of Huneke et al., 2022, they plot the bottom velocity (I think?), so perhaps that's the best thing to cluster? The only issue is I'm not sure how best to calculate the along-slope velocities for a range of isobaths, but I think you or @wghuneke would have the code to do that?

[wghuneke]

The calculation of the along-slope velocity does only depend on the local (grid cell) bathymetric slope and can be calculated for every grid point. So no need to extract any isobaths beforehand - in fact I calculate the along-slope component first and then select an isobath, if I remember correctly.

The along-slope velocity will have neg and pos values which might help in isolating what I call the "reverse ASC" (= ACC) in West Antarctica.

In general, I'd prefer an approach where we don't constrain the input for the GMM too much. Creating a mask based on bathymetry values makes sense, I think. E.g., Fig 2b in Huneke et al. 2022 shows the velocity for a mask shallower than 2500 m. Maybe 3000 m would be a good starting point. But if we start to select too narrow values/individual isobaths, then we make already lots of assumptions. My hope was that the GMM approach would (ideally) do exactly that for us.

Finally, Fig 5b shows the barotropic velocity, I think that's a better approach as the bottom velocity. The bottom will be large in the bottom-intensified regions (where there is DSW export), but close to zero elsewhere.

[wghuneke]

@taimoorsohail - do you want me to prep a file of the along-slope velocity? What time period + frequency did you use for your GMM analysis?

[taimoorsohail]

@wghuneke That would be amazing! I used the time-mean velocity in the final 10 years of the RYF run, so perhaps the monthly along-slope velocity in those final 10 years would be good?

I can calculate the time-mean myself, or otherwise explore seasonal/annual variability.

[wghuneke]

Just a note that we're not following up on the ASC definition anymore. We decided to use the isobath definition and potentially repeat the calculation for different isobaths.