[MITgcm-devel] netcdf units in diagnostics pkg

[Jean-Michel Campin]

Hi Martin,

On Thu, May 22, 2008 at 02:52:24PM +0200, Martin Losch wrote:
> Dear diagnostics-authors,
> 
> Sorry for bothering you with this, but I would like to understand the  
> logic of the routine body getdiag (diagnostics_utils.F, see below):

Courageous guy !

> 
> The first case (mate=0) happens (according to my observations, not my  
> understanding of the code, which I have to admit I don't understand)  
> whenever diagnostics_fract_fill has been used to fill the diagnostics  
> (e.g. SI_Thick), the second case happens when diagnostics_fill has  
> been used to fill the diagnostics (e.g., SI_Fract). In the first  
> case, any value above "undef" (hardwired to getcon('UNDEF') = 1D15)  
> is reset to "undef", in the second case "undef" is only used if the  
> diagnostics has been filled less than once (=not at all).
> 
> My questions:
> 1. In the above example (SI_Fract and SI_Thick), SI_Fract contains  
> zeros where there is no ice (iceMask=0), good!. SI_Thick, which  
> should be iceHeight*iceMask (and should be zero where iceMask=0),  

I disagree, at the end we want: 
SI_Thick = Sum( iceHeight*iceMask ) / Sum( iceMask )

> there are values of undef=getcon('UNDEF')=1D15 where there is no ice.  
> Now, whether this is useful or not, I just don't understand, where  
> these numbers come from (in getdiags, this value is only assigned if  
> the diagnostics exceeds "undef", or has not been filled).

See my comments in the code below.

> 2. is it a good idea to reset a diagnostic to "undef", when it is  
> larger than "undef", while "undef" is hardwired? As a counter  
> example: I can image averaging some pressure variable (e.g. bottom  
> pressure anomaly) over many timesteps (or maybe even the square of  
> bottom pressure) and would expect to get really large numbers that  
> are then capped by this logic.

I agree, this is not a good idea. But we need to accomodate with 
fizhi pkg that puts some undef on special occasions. 
If you want to change it, I would recommand something like 
 IF (useFizhi ) THEN / ELSE /ENDIF

> 
> Martin
> 
> >        IF ( mate.EQ.0 ) THEN
> >C-      No counter diagnostics => average = Sum / ndiag :
> >
> >          ipnt = ip + lev - 1
> >          factor = FLOAT(ndiag(ip,bi,bj))
> >          IF (ndiag(ip,bi,bj).NE.0) factor = 1. _d 0 / factor
> >
> >          DO j = 1,sNy+1
> >            DO i = 1,sNx+1
> >              IF ( qdiag(i,j,ipnt,bi,bj) .LE. undef ) THEN
> >                qtmp(i,j) = qdiag(i,j,ipnt,bi,bj)*factor
> >              ELSE
> >                qtmp(i,j) = undef
> >              ENDIF
> >            ENDDO
> >          ENDDO
> >
> >        ELSE
> >C-      With counter diagnostics => average = Sum / counter:
> >
> >          ipnt = ip + lev - 1
> >          levCt= MIN(lev,kdiag(mate))
> >          ipCt = im + levCt - 1
> >          DO j = 1,sNy+1
> >            DO i = 1,sNx+1
> >              IF ( qdiag(i,j,ipCt,bi,bj) .NE. 0. ) THEN
                      ^=== here is where Undef comes from when
       using a counter diag. Note the 3rd index = ipCt (counter) 
       which is not the same as "ipnt" 
> >                qtmp(i,j) = qdiag(i,j,ipnt,bi,bj)
> >     &                    / qdiag(i,j,ipCt,bi,bj)
> >              ELSE
> >                qtmp(i,j) = undef
> >              ENDIF
> >            ENDDO
> >          ENDDO
> >
> >        ENDIF
> >       ENDIF
> >
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel