Re: ferret 5.8 does things it should not do

[Ansley Manke <ansley.b.manke@xxxxxxxx>]

Hi Martin,
Thank you for writing.  We like to know how the changes we make to 
Ferret affect long-time users like you. Well, we make things more 
convenient for some, and unfortunately changing the output makes things 
difficult for others.  I'd first like to note that we have a project in 
the works now to give users complete control over netCDF attributes, 
which will make it easier to work with all of the attributes and control 
what is written to your output files. This may answer much of what 
you're asking for. I described a bit more about this project in a 
message not too long ago: 
http://www.ferret.noaa.gov/Ferret/Mail_Archives/fu_2005/msg00248.html

-- With the introduction of the concept of "subspan modulo" axes we set 
Ferret so that an axis with units of degrees longitude is modulo by 
default. This allows us to use different longitude axes together in 
comparisons. What kind of trouble is this causing you?  The Ferret 
command to remove the modulo behavior is

  cancel axis/modulo `var,return=xaxis`

-- The implementation of the bounds attribute is part of our move 
towards adherence to the CF metadata standard. 
(http://www.cgd.ucar.edu/cms/eaton/cf-metadata/CF-1.0.html ).  We could 
think about adding a qualifier SAVE/NOBOUNDS.  I'll have to think 
through whether there are implications of this that make it undesirable, 
but it seems like a simple solution.

-- First a quick explanation of the use of bounds when appending time 
series. Time axes have an upper bound, seen in the Ferret 
pseudo-variable TBOXHI[gt=var] and a lower bound, TBOXLO[gt=var].  When 
appending more time steps to a time axis, there will often be a gap 
between the last  upper bound of the existing data and the first lower 
bound of the new data.  If there is, then Ferret inserts the void 
point.  (If the new lower bound is less than or equal to the old upper 
bound, no void point is inserted.) We felt this was a more accurate way 
to represent the data.  For instance if a year's monthly data is output, 
January through December 2002, and then some data from the next year, 
April through November 2003 is appended, then a void point is added 
whose lower bound is the end of December 2002 and upper bound is the 
start of April 2003.  Without this void point, there is a data point 
with a large grid cell: the April 2003 data would seem to extend from 
the start of January to the end of April. 
One could imagine getting something similar to the old behavior by 
defining a time axis with the /EDGES qualifier to tell Ferret that you 
want the first lower bound of the data being appended to match the upper 
bound of the last time from the previous data.


Martin Schmidt wrote:

> Dear Ansley,
>
> I am using ferret 5.8,
>
> FERRET v5.80  Linux(g77) 2.4.20 - 01/03/05
> 22-Mar-05 16:17
>
> and found several points, where ferret "knows better" than it should. 
> Opening a netcdf-file
>
> use my_file
>
> and writing one of the variables with (doing nothing else)
>
> save/clobber/file=my_file.nc my_var
>
> I see, the x-axis is now modulo.        Originally:
>
> float xt_i(xt_i) ;
>                xt_i:long_name = "Longitude of T points" ;
>                xt_i:units = "degrees_E" ;
>                xt_i:cartesian_axis = "X" ;
> new file:
> double XT_I(XT_I) ;
>                XT_I:units = "degrees_east" ;
>                XT_I:modulo = 360. ;
>                XT_I:point_spacing = "uneven" ;
>                XT_I:axis = "X" ;
>                XT_I:bounds = "XT_I_bnds" ;
>
> I could cancel the modulo flag, but wouldn't it be better, if ferret 
> leaves the axis definition
> as it is? A previous version (5.4) did not add the modulo flag.
>
> It also adds always bounds which can be very disturbing in some cases. 
> This cannot be supressed.
>
> The final issue is:
> With previous ferrets one could concatenate files, which have the same 
> grid but an unlimited time axis. It worked simply by opening  file 
> after file and writing to an output file.
> This does not work anymore, because ferret adds void timeslices. 
> Indeed the time axis is irregular, but I did neither specify the 
> bounds attribute not used /rigid to get a regular axis. I could not 
> find out, how to get rid of these void points.
>
> Especially for the last point any idea for a workaround is welcome.
>
> Kind regards,
> Martin Schmidt