Closed PaulSpence closed 2 years ago
It looks like the Up scenario is leading to more upwelling (blue) on the shelf, and Down has more downwelling (red). I think this is due to the strong meridional wind forcing anomaly. The zonal wind components should be doing the opposite, e.g. stronger easterlies=more downwelling on shelf.
Is this wind perturbation doing what we want?
Yes, it's a bit hard to see with all the positive and negative values. I would expect that whatever sign of Ekman pumping is in the control will just be strengthened in the UP case. Perhaps regridding onto a 0.5deg or so grid would help? Have you also tried plotting the stress on the sea ice in the control to see how the direction of that compares with the UP anomaly?
Regarding, whether the wind perturbation is doing what we want: I would argue it is. When we set up the perturbation, we thought about how there is so much uncertainty in how the easterlies will change in the future, given that CMIP models can't reproduce anything like the observed winds, and there is also large uncertainty over recent decadal changes. So we decided that the cleanest perturbation would be to just amplify and weaken the existing winds. This doesn't mean that the simulations are a projection for the future, but hopefully we will gain a dynamical understanding of what the easterlies do in an idealised case. If you look at a map of the present day winds (JRA) along the coast (e.g. top plot here), you see that the easterlies aren't purely easterly at all, and that the meridional winds over the shelf are generally of the same magnitude as the zonal winds, and depending on the exact location, the meridional winds can be much larger. Here's a plot of the absolute value of the zonal and meridional winds to highlight this:
Is there actually upwelling in the UP case or just less downwelling (blue in the anomaly plot)? In any case it might be a hint to why we see warming in many regions of the continental shelf.
I presume you have some canned smoothers that you have used. But if you wish to use something that is gaining traction in the community, and which is under review at JAMES, then I recommend this smoother from Ian Grooms, Ryan Abernathey, and company. I have heard it works well "out of the box", which it should since they are trying to get the code published in JOSS.
Thanks Steve :) The figs below use the conservative regridding of xesmf, but I will check out the JAMES code this morning for fun. The Ekman pumping signal is still clear as mud :)
Figure 3 in this paper
https://www.sciencedirect.com/science/article/pii/S1463500318303779?via%3Dihub
shows w_curl = (1/rho0) curl (tau/f) from a suite of wind products, including JRA55-do. The patterns are rather smooth and interpretable. I wonder if we are computing the same diagnostic as per their equation (3), or are we computing (1/frho0) * curl (tau)?
I think we want to compute (1/rho0) * curl (tau/f) to quantify how much volume flux moves across the Ekman layer base. The patterns might not be that different whether we include 1/f inside or outside the curl, but worth having a check.
Good question Steve. It certainly was my intent to calculate Ekman as: w_curl = (1/rho0) * curl (tau/f). Though this time I was forced to upgrade my trusted curl technique to follow a cookbook recipe (method 3 from https://cosima-recipes.readthedocs.io/en/latest/documented_examples/RelativeVorticity.html#gallery-documented-examples-relativevorticity-ipynb)
My code is here: https://github.com/adele157/easterlies-collaborative-project/blob/master/notebooks/Ekman.ipynb
The relevant line is: we_con_xgcm = ( grid_con.interp( grid_con.diff(ds_con.tau_y/f, 'X') / grid_con.interp(ds_con.dxu, 'X'), 'Y', boundary='extend')
It is a good example of how python code can be quite obscure :)
I took a quick stab at reproducing Fig. 3 from your paper. The color scheme is reversed. I think your Fig. 3 looks so smooth because it has big (6 cm/day) colorblocks that smooth over features.
Ok, so the 1/f is inside the curl...that is good.
What causes the zonal strips in your figure?
While the Ekman pumping in the figures above looks very noisy around Antarctica, if you zoom in there is a lot more structure. Here's what it looks like in the Amundsen Sea for the whole 15 years of the perturbation runs:
The control Ekman pumping (and UP anomaly) is upward over nearly the whole region. This is consistent if you compare with the zonal wind stress: the easterlies keep increasing as you go south all the way to the coast. It's only right at the ocean/land boundary that the downwelling shows up.
The average over the continental shelf region shown here is -0.032 m/day for the control with anomalies of -0.002 m/day for the UP and 0.002 m/day for the DOWN perturbations.
@PaulSpence, I found a problem with your code. The xgcm method spreads nans at the ocean/land boundary, so you are missing most of the coastal downwelling in your plots. You can avoid this with some careful filling of masks, as I've done here.
Very nice images.
Thanks Adele for helping :) I've calculated curls in so many different languages (fortran, ferret, matlab, and 3 deprecations in python) that it kind of makes me want to puke :0
Next ...
P
On Fri, Aug 6, 2021 at 1:41 PM Stephen Griffies @.***> wrote:
Very nice images.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/adele157/easterlies-collaborative-project/issues/20#issuecomment-893977624, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABSWJXDYWQMOQUXM5VCTHLDT3NKVXANCNFSM5AMU2Q4A . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email .
-- Paul Spence, PhD ARC Future Fellow School of Geosciences University of Sydney, Australia https://paulspence.github.io/
@PaulSpence, yes it took me a while to figure out too. I wonder if we should put a warning in the cosima recipes curl example about this. Perhaps not such a big deal for the relative vorticity, but a bit of a killer for using the same method for Ekman pumping. @AndyHoggANU
I remember not being able to get the recipe method 2 working either (but have since deleted my attempt). Method 2 is a less obscure methodology.
P
On Fri, Aug 6, 2021 at 2:02 PM Adele Morrison @.***> wrote:
@PaulSpence https://github.com/PaulSpence, yes it took me a while to figure out too. I wonder if we should put a warning in the cosima recipes curl example https://github.com/COSIMA/cosima-recipes/blob/master/DocumentedExamples/RelativeVorticity.ipynb about this. Perhaps not such a big deal for the relative vorticity, but a bit of a killer for using the same method for Ekman pumping. @AndyHoggANU https://github.com/AndyHoggANU
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/adele157/easterlies-collaborative-project/issues/20#issuecomment-893983501, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABSWJXEBIIYTGTVXXTPJTILT3NNE5ANCNFSM5AMU2Q4A . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email .
-- Paul Spence, PhD ARC Future Fellow School of Geosciences University of Sydney, Australia https://paulspence.github.io/
Hi Adele,
I played with your Ekman pumping code today (thanks again for kindly digging in to help). I appreciate that your code fixes the xgcm calculation on ocean grid cells next land :). Otherwise the results are the same as the xgcm cookbook method 3. In particular, both methods leave some funny horizontal grid lines in the global plots (see attached figs). I am not sure what the issue is with these weird grid lines. They're not a big deal for our science, but maybe we can ask for some CMS help in resolving them? I will upload my code to the github repo for this calculation. Andy, maybe you could note this in the cookbook issue you created today?
Thank you, Paul
On Fri, Aug 6, 2021 at 2:51 PM P Spence @.***> wrote:
I remember not being able to get the recipe method 2 working either (but have since deleted my attempt). Method 2 is a less obscure methodology.
P
On Fri, Aug 6, 2021 at 2:02 PM Adele Morrison @.***> wrote:
@PaulSpence https://github.com/PaulSpence, yes it took me a while to figure out too. I wonder if we should put a warning in the cosima recipes curl example https://github.com/COSIMA/cosima-recipes/blob/master/DocumentedExamples/RelativeVorticity.ipynb about this. Perhaps not such a big deal for the relative vorticity, but a bit of a killer for using the same method for Ekman pumping. @AndyHoggANU https://github.com/AndyHoggANU
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/adele157/easterlies-collaborative-project/issues/20#issuecomment-893983501, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABSWJXEBIIYTGTVXXTPJTILT3NNE5ANCNFSM5AMU2Q4A . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email .
-- Paul Spence, PhD ARC Future Fellow School of Geosciences University of Sydney, Australia https://paulspence.github.io/
-- Paul Spence, PhD ARC Future Fellow School of Geosciences University of Sydney, Australia https://paulspence.github.io/
It would be good to plot more zoomed in Ekman pumping plots like the Amundsen ones above. Do we see the same coastal downwelling / widespread upwelling in the UP simulation in other regions? We could also sum the Ekman pumping over coastal (perhaps everywhere less than ~200m depth?) / non coastal regions to smooth some of the noise, and distinguish the coastal / non-coastal responses.
At least there is overall downwelling on the shelf :)
Very nice figures.
--any smoothing? --are you confident with the values next to coast? In some images the coast is where most of the signal is maxed, so wish to be sure of the land treatment.
Calc Ekman pumping as function of outcropping rho class across the shelf (check the watemass transformation code and Kial for code).
Hi Steve,
No smoothing applied in those figures.
Curls next to the coast ... we had a brief discussion on this today. I was wondering if taking a derivative with a land value taux=NaN is sensible ... Adele and Andy suggested it was. In this case the land value taux=0 (option 2 code), whereas other codes NaN the values next to the coast.
Thanks for the comments.
Paul
On Thu, Sep 9, 2021 at 9:47 AM Stephen Griffies @.***> wrote:
Very nice figures.
--any smoothing? --are you confident with the values next to coast? In some images the coast is where most of the signal is maxed, so wish to be sure of the land treatment.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/adele157/easterlies-collaborative-project/issues/20#issuecomment-915646856, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABSWJXDD2QX3NODTAKBOIB3UA7YYNANCNFSM5AMU2Q4A . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
-- 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/
I think taux=0 is the right method for land values, not taux=NaN. Otherwise you'll miss out on the downward Ekman pumping right next to the coast. At the coast we have southward Ekman transport and nowhere else for that water to go but down into the water column, I think that's real dynamics not an artefact.
I concur that tau=0 over land is proper choice. That's generally how the model computes shear next to land.
Agreed team. My question today about the near coastal pumping anomalies was just making sure the curl there was being computed correctly right at the coast - i.e. using zero values over land (even though as I understand it JRA-55 will have a non-zero wind value there over land). Also, when we apply the +/- 10% it's still just \tau = 0 at the coast for this pumping computation. Thanks for confirming Adele & Steve.
Yes, I can confirm that Paul's regional plots above use the correct tau = 0 over land in the calculation. The original circumpolar plots right at the top of this issue used tau = NaN, but we've corrected the code now.
And yes Matt, for all the perturbations, we're also using tau = 0 over land.
I presume you have some canned smoothers that you have used. But if you wish to use something that is gaining traction in the community, and which is under review at JAMES, then I recommend this smoother from Ian Grooms, Ryan Abernathey, and company. I have heard it works well "out of the box", which it should since they are trying to get the code published in JOSS.
FYI gcm_filters
is now available in the conda/analysis3-unstable
on NCI.
Thanks @aidanheerdegen
Agreed team. My question today about the near coastal pumping anomalies was just making sure the curl there was being computed correctly right at the coast - i.e. using zero values over land (even though as I understand it JRA-55 will have a non-zero wind value there over land). Also, when we apply the +/- 10% it's still just \tau = 0 at the coast for this pumping computation. Thanks for confirming Adele & Steve.
@matthew-england-unsw uses GitHub?!!!
Good question Steve. It certainly was my intent to calculate Ekman as: w_curl = (1/rho0) * curl (tau/f). Though this time I was forced to upgrade my trusted curl technique to follow a cookbook recipe (method 3 from https://cosima-recipes.readthedocs.io/en/latest/documented_examples/RelativeVorticity.html#gallery-documented-examples-relativevorticity-ipynb)
My code is here: https://github.com/adele157/easterlies-collaborative-project/blob/master/notebooks/Ekman.ipynb
This notebook doesn't exist.....
Time-series of annual mean, area averaged Ekman Pumping on Shelf in (almost) all experiments. Don't worry about end of green line :)
Spatial distribution of Control state annual mean Ekman pumping and sigma0 isopycnal contours (for binning of pumping in surface density classes)
Spatial distribution of Control state Summer/Winter sigma0 isopycnals (for binning)
Ekman pumping integrated south of wind perturbation line in Julia's Fig. 1, and on/off shelf.
Ekman pumping in sigma classes in Amundsen Sea region
Density and isotherms in Amundsen Sea region
Ekman pumping in sigma classes in Denman region
Density and isotherms in Denman region
Ekman pumping in sigma classes in Weddell Sea
Density and isotherms in Weddell Sea
Nice! Yes, I would say even 27.75-27.8 for the minimum CDW density offshore? The problem then, is that those densities are actually downwelling, not upwelling in the Ekman pumping plots.
@julia-neme and I were chatting this morning and she raised the idea that maybe these CDW isopycnals don't actually outcrop much, i.e. the transformation from CDW to AABW may occur mostly below the surface in overflow mixing regions, or otherwise only outcrop in small polynya regions near the coast. This does seem to be what the Ekman pumping plots are showing.
If that's true, where does this leave us? Do we still propose the Ekman pumping hypothesis, but then show that the relevant isopycnal range doesn't actually outcrop or experience Ekman-driven upwelling, and therefore rule that hypothesis out again straight away? Even though this is a simpler explanation of why this hypothesis doesn't work, I think it's still interesting, as it is very different from the idealised results of Stewart and Thompson 2012, who showed that the lower cell responded nearly linearly with the easterlies, following the expected Ekman change.
Re: "the transformation from CDW to AABW may occur mostly below the surface in overflow mixing regions, or otherwise only outcrop in small polynya regions near the coast. This does seem to be what the Ekman pumping plots are showing."
I am a little uncertain about the expectation that CDW density (or other deeper classes) need to appear cleanly in surface density in order for them to be outcropping. There are a lot of strong diabatic processes happening at the surface.
Density structure following along 1000m isobath (attempt at Fig. 5):