[MITgcm-devel] vectorization of layers

[Martin Losch]

Great, thanks for the clarification about the layers-depth.

Let's wait with the check-in until you've tried the code yourself, there is no rush.

Martin

On Jan 8, 2013, at 5:11 PM, Ryan Abernathey <ryan.abernathey at gmail.com> wrote:

> Martin,
> 
> Thanks so much for your work to improve this. I am traveling at the moment and have not been able to try out the code. I will be back on my office on thursday and can try it out then. It certainly *looks* fine. If you are convinced it is working well, I would go ahead and check it in.
> 
> Regarding the "mean depth", what I always do is to use the thickness variables to reconstruct the mean depth. For example, Hs gives the mean layer thickness at the v points. You just cumsum it up to give the layer boundary depths and then find the midpoints for the mean depths. This would only not work is when the water column is not stably stratified with respect for the tracer. For my purposes, this doesn't happen, but it would definitely happen with, for example, salinity. However, in this case, the mean depth of a tracer layer itself is poorly defined. What if there are two salinity maxima with the value of 35 psu at very different depths? Does it make sense to talk about mean depth of the 35 psu surface? So I guess my point is that I don't think it makes sense to add another variable for the mean depth.
> 
> I definitely think we are at the point where we need a more robust verification experiment.
> 
> -Ryan
> 
> 
> 
> 
> On Tue, Jan 8, 2013 at 9:54 AM, Martin Losch <Martin.Losch at awi.de> wrote:
> Hi Ryan,
> 
> did you get a chance to look at the changed layers_fluxcalc.F?
> I also tried to further the vectorization and I did succeed technically, but at the cost of nearly infinite execution time, so even if my idea gives the correct results, there is no computational gain (although nicely vectorized). I feel that, what we have now (what I sent with my last email), is the best I can do right now. I am happy to check that in, if you approve of it.
> 
> I have a general question. Does it make sense to compute (and save) the "mean depth" of the layers? I would need that to compute, e.g. the overturning stream function averaged on isopycnals but plotted as a function of latitude and depth. Reconstructing the depths after the run does work, but it becomes a little noisy (as does the post-mortem computation of the overtuning on density surface, that's why your layers package is so useful).
> 
> Martin
> 
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel