[MITgcm-devel] exf_inscal_u/vwind

[Martin Losch]

Hi Jean-Michel,

this looks good to me, i.e. moving to before vector rotation (l226 in exf_set_uv.F). Interpolation to the grid should be done first, I agree with that too.

Martin

> On 09 Sep 2016, at 09:30, Jean-Michel Campin <jmc at mit.edu> wrote:
> 
> Hi Martin,
> 
> I (finaly) took a look at the code, and I would suggest an intermediate solution
> regarding where exf_inscal_u/vwind is applied:
> I would like to move it before rotation to model grid but still after S.pole & N.pole
> treatment. This will enable to make changes later to add back time dependent factor
> (like uwind_exfremo_slope) for each component before rotation (but I don't plan
> to do it now). 
> 
> And regarding the interpolation near the S & N poles, I am not sure of 
> what make more sense, so it's probably not too bad; but good to keep in mind
> when inscal_u & inscal_v are verry different.
> 
> Does this look good ?
> 
> Cheers,
> Jean-Michel
> 
> On Mon, Aug 15, 2016 at 09:01:49AM +0200, Martin Losch wrote:
>> Hi Jean-Michel,
>> 
>> thanks for looking into that. It???s really only interesting in some idealized cases, but since the other fields behave in that way (just because of the missing interpolation, the scaling factor is actually applied at the same place as in exf_set_uv.F), it may be more consistent to scale the input fields.
>> 
>> I don???t know about the other two parameters, never used them ???
>> 
>> Martin
>> 
>>> On 12 Aug 2016, at 22:17, Jean-Michel Campin <jmc at mit.edu> wrote:
>>> 
>>> Hi Martin,
>>> 
>>> I think I contribute a little bit in putting some STOP to prevent interpolation of
>>> winds when they are not provided together and at the same time. Since there was
>>> some previous wind interpolation code (to do each comp. separately) and still in
>>> used for simple grid case, I though it would be more safe. 
>>> So yes, you might need to provide a file full of zeros for that.
>>> 
>>> Regarding exf_inscal_u/vwind: I did mot think about this scaling factor, and
>>> agree with you that the scaling should apply before 2-component interpolation 
>>> and not after. Will check if it's easy to change (unless others disagree).
>>> 
>>> I still have issues with these 2:
>>>> uwind_exfremo_intercept, uwind_exfremo_slope
>>> (note that currently, they are not implemented in exf_set_uv.F for the case
>>> #defined USE_EXF_INTERPOLATION)
>>> a) more difficult to apply time-dependent scaling ion each comp. before interp.
>>>   since interp is applied once, just after loading files, and not at each time-step.
>>> b) this factor: XXX_remove_slope*(myTime-startTime)
>>>   will always break the restart test (intentional ?). 
>>> 
>>> Cheers,
>>> Jean-Michel
>>> 
>>> On Thu, Aug 11, 2016 at 09:05:45AM +0200, Martin Losch wrote:
>>>> Hi Jean-Michel,
>>>> 
>>>> I observed something unexpected during debugging my llc90-configuration. I want to run the model with (geographically) zonal wind only (exf/uwindfile with interpolation and rotation to the grid). When I set vwindfile = ??? ???, the model complains in exf_check that it need two components, OK. Then I tried exf_inscal_vwind = 0, because I thought that this would scale the ???in-field??? and that I then would end up with the zonal wind distributed between uwind and vwind appropriately on the faces. But in fact exf_inscal_u/vwind scale the model field uwind and vwind, so that I end up with no vwind and a mix of zonal and meridional wind on ???uwind???, as usual.
>>>> 
>>>> Is this the intended behavior? The only way I see to run with only zonal wind is to have meridional wind file containing only zeros. 
>>>> 
>>>> 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
>> 
>> 
>> _______________________________________________
>> 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