[MITgcm-devel] Re: [MITgcm-cvs] MITgcm/pkg/diagnostics CVS Commit

[Ed Hill]

On Tue, 2006-02-07 at 10:25 -0500, Andrea Molod wrote:
> hi baylor,
> 
> > I suggest that missing value shouldn't be zero by default.  That just 
> > confirms the old problem where masked fields were set to zero in the masks. 
> > For fields like temperature and velocity that was just a mess.  If we use 
> > missing_value=0, and open in ferret or netcdf, then, for example, in a run 
> > started from rest ALL the initial velocities will be missing!  I suggest the 
> > missing value standard from netcdf:  -1e34.
> 
> 
> i agree that 0. is a potentially problematic missing value. (i have
> been using 1.e15 in some parts of the code and converting back 
> and forth) but the issue is not only for output files, it is for the model 
> itself (asking the output routines to put -1e34 into the places where the
> field is 0.0 doesn't help much). until we discuss and resolve this, i
> asked ed to add the value we are all using anyway (and is in the fields 
> themselves!) as a netcdf file attribute. i hope that we can leave it
> this way until we deal with (or decide that 0. is ok) it in a somewhat
> consistent manner.

Hi Andrea & Baylor,

Per Baylor's observations, I've commented out the missing_value
attribute.  This should cause the least overall disruption since it will
not break anything for Baylor (his ferret scripts) and Andrea can easily
uncomment the two lines in diagnostics_out.F for her work.

Looking ahead, lets discuss what ought to be done here in order to make
everything work smoothly.  At a min, it seems that the diagnostics code
needs a clear-cut (that is, completely unambiguous) way of marking
missing values.  Other groups use "obviously" large or small values such
as 1.e15, etc.  And I think we'll want to do it on a per-variable basis
with the ability for users to optionally specify what these per-variable
missing values are so that they're out-of-range from "normal" values.

Are there any other issues or ideas that we ought to discuss before we
implement this?  For instance, many netCDF tools (eg. NCO) support the
concept of one or more "mask" variables.  Would it smarter (more
efficient) for us to use a small number of mask variables rather than
effectively embedding masks in a large number of fields using the
missing_value feature?

Ed

-- 
Edward H. Hill III, PhD
office:  MIT Dept. of EAPS;  Rm 54-1424;  77 Massachusetts Ave.
             Cambridge, MA 02139-4307
emails:  eh3 at mit.edu                ed at eh3.com
URLs:    http://web.mit.edu/eh3/    http://eh3.com/
phone:   617-253-0098
fax:     617-253-4464