[MITgcm-support] exf interp

Martin Losch Martin.Losch at awi.de
Mon Oct 19 03:37:49 EDT 2009 

Sorry, I did not read the thread carefully. Please disregard my  
superfluous comments.

Martin

On Oct 19, 2009, at 9:28 AM, Martin Losch wrote:

> Hi Matt,
>
> I don't know if this helps, but the wind stress is something like  
> Cd*sqrt(u^2+v^2)*(u,v). Then the wind stress curl is A*dv/dx - B*du/ 
> y + C*du/dx - D*dv/y with A-D some simple function of Cd,u,v  
> (basically u^2/sqrt(u^2+v^2).
>
> Now if you your wind field is piece-wise linear (because of the  
> linear interpolation between the 1-deg grid points), then your du/ 
> dx,dv/dx,du/dy,dv/dy are piecewise constant. In other words they  
> have discontinuities at cell interfaces, which explains your  
> pattern. This pattern can only be eliminated with a higher order  
> interpolation scheme (as Chris points out at least c1 continuous, so  
> that the deriviative is continuous, ie, at least piecewise linear).  
> We should consider making bicubic default (only) for wind  
> interpolation.
>
> Martin
>
> On Oct 17, 2009, at 11:53 PM, Dimitris Menemenlis wrote:
>
>> The issue was mainly one of negative rain
>> (and perhaps also negative downward radiations and humidity).
>> These will probably not be a problem in your limited domain.
>>
>> When you run your test on Monday, could you also save the wind  
>> velocity fields:
>>
>> 179 |EXFuwind|  1 |SM      U1      |m/s             |zonal 10-m  
>> wind speed, >0 increases uVel
>> 180 |EXFvwind|  1 |SM      U1      |m/s             |meridional 10- 
>> m wind speed, >0 increases uVel
>> 181 |EXFwspee|  1 |SM      U1      |m/s             |10-m wind  
>> speed modulus ( >= 0 )
>>
>> to see if there are any discontinuities are in the curl of the wind  
>> velocity.
>> If Chris' hypothesis is correct you will not see a rectangular  
>> pattern in
>> the curl of the wind.  if you do, then we will need to use Chris'
>> interpolation scheme.
>>
>> D.
>>
>> On Oct 17, 2009, at 1:38 PM, Matthew Mazloff wrote:
>>
>>> Hi Chris,
>>>
>>> Yeah, its relatively easy to test.  I'll give it a try on monday and
>>> let you know
>>>
>>> -Matt
>>>
>>> Dimitris, what issues arose when using bicubic before for buoyancy
>>> terms -- anything specific I should look out for?
>>>
>>> -Matt
>>>
>>>
>>> On Oct 17, 2009, at 1:14 PM, Chris Hill wrote:
>>>
>>>> Matt,
>>>>
>>>> I think the bicubic is worth trying. As I understand Benny's code  
>>>> it
>>>> is meant to be c1 continuous
>>>> which should help.
>>>> My hope is that the bilinear temps are affecting the wind stress
>>>> (because atemp/SST feature in static stability calcs in bulk  
>>>> formula
>>>> and in turbulent mixing calcs and so affect momentum flux).
>>>> Is it quick(ish) to run a test?
>>>>
>>>> If this doesn't work I have a couple of other things that would be
>>>> kind of interesting to try. They involve doing stuff offline
>>>> initially.
>>>>
>>>> Chris
>>>>
>>>> On Sat, Oct 17, 2009 at 1:36 PM, Matthew Mazloff <mmazloff at mit.edu>
>>>> wrote:
>>>>> Hi Chris,
>>>>>
>>>>>> Are the surface data sets 1/2 degree res?
>>>>>
>>>>> 1 degree res -- starting at the half degree:
>>>>> uwind_lon0         = 229.5D0,
>>>>> uwind_lon_inc      = 1.D0,
>>>>> uwind_lat0         = 26.5D0,
>>>>> uwind_lat_inc      = 15*1.D0,
>>>>>
>>>>>> My guess is it is the piecewise constant in the aTemp/aQ/ 
>>>>>> radiation?
>>>>>
>>>>> I'm confused, I'm plotting windstress curl, and the wind speed  
>>>>> uses
>>>>> bicubic
>>>>> interpolation already.
>>>>> Do you think this is a feedback from buoyancy forcing be linearly
>>>>> interploated?  I can check the wind speed to ensure the signal is
>>>>> not coming
>>>>> from the (ocean) relative speed in the stress calculation.
>>>>>
>>>>>> If the spline is done right
>>>>>> the overshoots should be small in a limited area domain, so you
>>>>>> could try
>>>>>> that.
>>>>>
>>>>> You mean trying bicubic for buoyancy components?
>>>>>
>>>>>
>>>>> Thanks for the help!
>>>>> -Matt
>>>>>
>>>>>
>>>>>
>>>>>>
>>>>>> Chris
>>>>>>
>>>>>> On Sat, Oct 17, 2009 at 11:38 AM, Matthew Mazloff <mmazloff at mit.edu 
>>>>>> >
>>>>>> wrote:
>>>>>>>
>>>>>>> Hi Dimitris,
>>>>>>>
>>>>>>> See the lines at X.5 degrees.  Its not very noticeable in
>>>>>>> snapshots, but
>>>>>>> really stands out in the mean as the signal is accumulated
>>>>>>>
>>>>>>> -Matt
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> On Oct 17, 2009, at 7:42 AM, Dimitris Menemenlis wrote:
>>>>>>>
>>>>>>>> Matt, could you send an example figure of problem?
>>>>>>>>
>>>>>>>> On Oct 17, 2009, at 7:29 AM, Matthew Mazloff wrote:
>>>>>>>>
>>>>>>>>> Hi Dimitris,
>>>>>>>>>
>>>>>>>>> Ah yes, I see this now.  Very nice.  Unfortunately the  
>>>>>>>>> gridding
>>>>>>>>> is
>>>>>>>>> still noticeable in my plots of wind speed gradient.  I am  
>>>>>>>>> able
>>>>>>>>> to
>>>>>>>>> pick out the exact locals where the forcing is prescribed --
>>>>>>>>> meaning
>>>>>>>>> the interpolation is not smooth.  This is not good for  
>>>>>>>>> plotting
>>>>>>>>> wind-
>>>>>>>>> stress curl. Have you noticed this on your high-res set-ups?
>>>>>>>>> Do you
>>>>>>>>> think this is a problem?  Should we (can we) try a smoother
>>>>>>>>> interp
>>>>>>>>> method?
>>>>>>>>>
>>>>>>>>> -Matt
>>>>>>>>>
>>>>>>>>> On Oct 16, 2009, at 5:03 PM, Dimitris Menemenlis wrote:
>>>>>>>>>
>>>>>>>>>> Matt, I hardcoded bilinear interpolation for tracer fields
>>>>>>>>>> because
>>>>>>>>>> overshoots can be problematic and bicubic for wind velocity  
>>>>>>>>>> (see
>>>>>>>>>> exf_set_uv.F) for stress fields because the second derivative
>>>>>>>>>> matters.  I would recommend to leave as is.  Other
>>>>>>>>>> combinations can
>>>>>>>>>> (or did) cause trouble.  Dimitris
>>>>>>>>>>
>>>>>>>>>> On Oct 16, 2009, at 3:21 PM, Matthew Mazloff wrote:
>>>>>>>>>>
>>>>>>>>>>> Hello,
>>>>>>>>>>>
>>>>>>>>>>> It appears the model is equipped to do both bilinear and
>>>>>>>>>>> bicubic
>>>>>>>>>>> interpolation for exf_interp.F.  Interp_method, however, is
>>>>>>>>>>> hardcoded
>>>>>>>>>>> to bilinear interpolation.  This does matter for my 1/16
>>>>>>>>>>> degree set-
>>>>>>>>>>> up.  Can anyone confirm that the bicubic interpolation is ok
>>>>>>>>>>> -- and I
>>>>>>>>>>> can go ahead and use this option.
>>>>>>>>>>>
>>>>>>>>>>> -Thanks,
>>>>>>>>>>> Matt
>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> MITgcm-support mailing list
>>>>>>>> MITgcm-support at mitgcm.org
>>>>>>>> http://mitgcm.org/mailman/listinfo/mitgcm-support
>>>>>>>
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> MITgcm-support mailing list
>>>>>>> MITgcm-support at mitgcm.org
>>>>>>> http://mitgcm.org/mailman/listinfo/mitgcm-support
>>>>>>>
>>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> MITgcm-support mailing list
>>>>>> MITgcm-support at mitgcm.org
>>>>>> http://mitgcm.org/mailman/listinfo/mitgcm-support
>>>>>
>>>>> _______________________________________________
>>>>> MITgcm-support mailing list
>>>>> MITgcm-support at mitgcm.org
>>>>> http://mitgcm.org/mailman/listinfo/mitgcm-support
>>>>>
>>>>
>>>> _______________________________________________
>>>> MITgcm-support mailing list
>>>> MITgcm-support at mitgcm.org
>>>> http://mitgcm.org/mailman/listinfo/mitgcm-support
>>>
>>> _______________________________________________
>>> MITgcm-support mailing list
>>> MITgcm-support at mitgcm.org
>>> http://mitgcm.org/mailman/listinfo/mitgcm-support
>>
>> _______________________________________________
>> MITgcm-support mailing list
>> MITgcm-support at mitgcm.org
>> http://mitgcm.org/mailman/listinfo/mitgcm-support
>
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support

More information about the MITgcm-support mailing list