[MITgcm-devel] smoothing package / penalizing controls

[Matthew Mazloff]

Hi Holly,

You want to penalize the pre-smoothed ctrls -- as our error field is  
only a variance.  The smoothed field represents the field with  
covariance.

Yes, the field is non-dimensionalized, so your error choice must be  
consistent with this.  You just need to adjust your uncertainty  
(weight).

-Matt




On Aug 10, 2011, at 11:21 AM, Holly Dail wrote:

> Hello -
>
> It seems to me that there is a problem  in how adjustments are  
> penalized when using the smoothing package.  I'm interested to know  
> if anyone has insight into this ... perhaps I'm missing something.
>
> Using theta0 adjustments as an example and assuming  
> ALLOW_SMOOTH_CORREL3D and ALLOW_SMOOTH3D are both on:
>
> - ctrl_map_ini_gen.F
> 		- loads the xx_theta file
> 		- calls smooth_correl3D on the field,
> 		- sets xx_theta = xxtheta/sqrt(weight)
> 		- modifies initial theta by the new xx_theta field
>
> - cost_theta0.F very simply loads the xx_theta file and computes  
> cost based on xx_theta.*xx_theta (no weights used)
>
> I would think cost_theta0 should use the xx_theta field that results  
> after smooth_correl3D and after dimensionalization.  I've been  
> writing out that modified field as xx_theta.effective2 for some time  
> now for analysis.  The easiest fix seems to be to use that field in  
> cost_theta0, instead of using xx_theta, but I'd be glad to hear of  
> any counterarguments that the current code is correct.
>
> Thanks -
> Holly
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel