[MITgcm-support] KPP diffusivities
Stefano Querin
squerin at ogs.trieste.it
Thu Nov 10 11:58:00 EST 2005
Hi,
I'm forwarding to the list a conversation I had with Dimitris about eddy viscosity/diffusivity and the KPP implementation on the MITgcm model.
Maybe someone could be interested or has some comments/hints/guesses about these topics.
Cheers,
Stefano
----- Original Message -----
From: Stefano Querin
To: menemenlis at jpl.nasa.gov
Sent: Wednesday, November 09, 2005 6:55 PM
Subject: Questions on KPP
S:
Hi Dimitris,
I have some other doubts about the KPP algorithm and its implementation on my case study [...] :
- I'm reading the paper by Large, McWilliams, Doney (1994, Rev. Geophys.) to find any possible explanation to my "problems". I also found very interesting the paper by Skyllingstad, Smyth, Crawford (Resonant Wind-Driven Mixing in the Ocean Boundary Layer, 2000, JPO). Is there any reference in which the implementation of KPP for the MITgcm model is described in detail? I would like to go a bit deeper into this topic.
D:
No except for some notes in MITgcm manual (which I have not yet read).
The KPP code in MITgcm (at least the part that is usually turned on) is
derived directly from NCOM so it should be no different than that
described in paper by Large et al. 1994 (except for some missing bells
and whistles, e.g., double diffusivity parameterization).
S:
Unfortunately, the KPP section in the MITgcm manual is empty (http://mitgcm.org/pelican/online_documents/node213.html). The NCOM users' manual explicitly refers to the paper by Large et al. (in particular to appendix D, for the numerical implementation) so that paper is definitely the most useful.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
S:
- as you already told me, Large in his paper says that KPP results are not very resolution dependent (he made a high resolution run (LOTUS experiment) with deltaz=0.50 m). That's encouraging!
- regarding your guess in the last mail: the plot I sent in my first mail (T-diffKzT.PNG) shows the time-depth profiles in a point in the middle of the basin, where, as you suspected, the stratification is weak. My doubts are caused by these observations:
a.. I found high bottom diffusivities almost in all the gulf, also where stratification is pretty stable;
b.. I found them also after many days of spin up (they are still present at the end of 20-days runs, after strong meteorological events, river floods...!)
c.. the diffusivity peaks have exactly the same shape of the surface sw flux (they are smaller during rainy days and so on...). I wonder how can surface irradiance cause those velocity perturbations and, therefore, those instabilities at the bottom;
D:
Is surface irradiance uniform? If it is non-uniform, wouldn't that
generate some currents? If irradiance is uniform but bathymetry is
non-uniform, wouldn't that also generate currents? I don't know if
these thermally driven currents would be significant.
S:
Yes, the irradiance is spatially uniform and the bathymetry is not, so the irradiance could generate currents. At first sight they seem absent or negligible (probably they are masked out by the wind driven currents...): I will examine velocities more in detail.
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
S:
a.. also the order of magnitude of the peaks is puzzling: values at the bottom are only slightly less than half the values on the surface, which are caused by very strong wind forcing (hourly mean values higher than 15 m/s!);
D:
Values for shear instability and for surface boundary layer mixing are
set, respectively, by
> c difm0 = viscosity max due to shear instability (m^2/s)
> c difs0 = tracer diffusivity .. (m^2/s)
> c dift0 = heat diffusivity .. (m^2/s)
> c difmcon = viscosity due to convective instability (m^2/s)
> c difscon = tracer diffusivity .. (m^2/s)
> c diftcon = heat diffusivity .. (m^2/s)
which default to
> difm0 = 0.005
> difs0 = 0.005
> dift0 = 0.005
> difmcon = 0.1
> difscon = 0.1
> diftcon = 0.1
For shear instability, the transition from no mixing to mixing is quite
abrupt. So it is essentially 0 or .005. For PBL, the values are scaled
with depth of mixed layer, with distance from surface, etc. So it is
possible for shear instability and PBL KPP viscosities to be of similar
magnitude.
S:
You're right. The fact is that I thought (erroneously) that the higher eddy viscosity/diffusivity coefficients you have, the more intense turbulent activity you obtain. That is not always true. The Boussinesq's eddy-viscosity concept says:
so, to have intense turbulence, you also need strong velocity gradients (of course...). In my case, those gradients are strong on the surface during the wind event and weak at the bottom during all the simulation.
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
S:
a.. disabling the KPP_SHEAR_MIX code, I found more reasonable valus for the diffusivity (nightime surface increase due to surface cooling...). I plotted 3 different tests (the variable is KPPdiffKzT, only one day of simulation) in the noslip_noshear_freeslip.PNG picture: the first is the "standard" configuration (top), the second is the one without the KPP shear mixing (center), the third is made using the same conditions as the first except for the use of free-slip conditions at the bottom. The most significant difference is in the order of magnitude of the diffusivities of the second plot (max 0.00055 vs. 0.01 of the other two). I also expected bigger differences between the first and the third: shouldn't I obtain much less shear at the bottom removing the zero velocity condition at the walls (free-slip)?
D:
In your data file you have "bottomDragQuadratic=0.001", which is added to
no-slip or free slip condition, and may be a larger term than the
no-slip condition, depending on velocities and on background viscosity
coefficient. It is not clear a priori which whill produce less shear:
free slip or no slip. All else being the same, smaller bottom drags
will cause larger velocities and therefore possibly larger shear?
S:
The data file I attached was used for the no_slip run (the first one, top in the figure). For the free_slip case I set:
no_slip_sides=.FALSE.,
no_slip_bottom=.FALSE.,
and I commented out
# bottomDragQuadratic=0.001,
Then the SET_DEFAULTS routine should have set the drag coefficients to zero.
Anyway, maybe larger velocities can cause larger shear: also in this case a deeper look in the velocity and shear profiles should be helpful.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
S:
My final question is:
a.. are those peaks in the bottom diffusivities physical? If not, should I modify some parameters (critical Ri...) in the KPP routines to avoid them?
D:
I don't know. A good representation of bottom PBL is sorely lacking in
MITgcm. It may be a good idea to transfer this conversation back to
MITgcm support list and see if anyone else, maybe Alistair Adcroft, can
comment.
S:
In fact the paper by Large at al. deals only with surface BL and ocean interior: there is no mention to bottom BL. Also the MITgcm implementation only sets eddy coefficients to zero at the bottom: bottom BL is not treated as a "physically different" region, as the surface BL, am I right?
D:
Yes. In standard configurations there is no special treatment of bottom
PBL, except for the bottomdrag quadratic or bottomdraglinear flags. But
someone else may have experimented with more complete bottom PBL
representations in some corner of MITgcm CVS server that I am not aware of.
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
S:
- I also think that there is something wrong in the surface diffusivities and in the T and S mixing dynamics.
In the zipped file I put the results of two 20-days simulations (spring situation (25/04 - 15/05) and summer situation (19/06 - 09/07): T and S in depth and time).
The plots on the top show the values measured by buoys (MAMBO 3 for the spring case, MAMBO 1 for the summer case, see the domain.png picture for the details), plots on the bottom are the model results in the correspondent positions.
Spring test: salinity comparison seems good, the low salinity signal is due to a flood of the Isonzo river. On the contrary it seems that there is a very weak mixing of temperature (maybe part of the problem is due to OBCs, the buoy mooring is very near to the open boundary of the model...).
Summer test: in this case, both T and S plots seem to show poor turbulent mixing, while the breaking (25/06) of the initial stable stratification due to wind driven mixing combined with coastal upwelling of bottom cold and salty water is well simulated by the model.
My question is:
a.. to obtain a better match with experimantal observations (i.e. enhance the pycnocline deepening) should I try different eddy parameters in PARM01 in the data file (attached)? I reasonably think that ICs and surface BCs are reliable and fit for my model accuracy (I made several checks on them).
D:
What about lateral BCs?
S:
I use no slip conditions for the solid walls and OBCs at the western (nesting) and northern (river) OB. Briefly:
a.. west: the model is nested into a POM model of the northern Adriatic sea. I used the useOBCSsponge option of the OBCS pkg (see the data.obcs file for details);
b.. north: I imposed river velocity, T and S as measured by a station placed in the riverbed, few kilometers from the mouth.
D:
The vertical mixing parameters I would attempt
to calibrate are diffKzT, diffKzS, and viscAz in PARM01 in "data" and
Ricr and Riinfty in data.kpp. A methodology for doing so in a
semi-organized way is given in following paper:
http://ecco.jpl.nasa.gov/~dimitri/articles/MWR05.pdf
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20051110/ddc15626/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: image/png
Size: 14902 bytes
Desc: not available
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20051110/ddc15626/attachment.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: T-diffKzT.PNG
Type: image/png
Size: 30690 bytes
Desc: not available
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20051110/ddc15626/attachment-0001.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: comparisons.zip
Type: application/x-zip-compressed
Size: 228351 bytes
Desc: not available
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20051110/ddc15626/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: noslip_noshear_freeslip.PNG
Type: image/png
Size: 73084 bytes
Desc: not available
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20051110/ddc15626/attachment-0002.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: domain.png
Type: image/png
Size: 124909 bytes
Desc: not available
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20051110/ddc15626/attachment-0003.png>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: data.txt
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20051110/ddc15626/attachment.txt>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: data.obcs.txt
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20051110/ddc15626/attachment-0001.txt>
More information about the MITgcm-support
mailing list