Dear Jon,
An addition to this would also be a theoretical max and min. Where
theory indicates that all valid values lie within a specific range we
could add this to the attributes. I.e. probabilities lying between 0 and
1 etc.
Kind regards
Julian
On Mon, 2007-01-08 at 10:31 +0000, Jon Blower wrote:
> Dear all,
>
> I think there are at least two types of min/max attributes that one
> could associate with a particular variable in a netCDF file:
>
> 1) The *actual* min and max of the variable's values in that
> particular file (not including missing values). Perhaps stored as
> data_min and data_max? Useful for both visualisation and data mining
> (as Bryan pointed out).
>
> 2) If the netCDF file is part of a larger dataset (e.g. it represents
> one timestep in a forecast sequence) it would be useful to store a
> "reasonable" (or actual, if possible) min and max for the *whole
> dataset* (taking into account the measurement method used,
> geographical region etc). Perhaps stored as dataset_min and
> dataset_max?
>
> The latter pair of attributes would really help when creating
> visualisations of timeseries across multiple netCDF files and could be
> a "soft" range: i.e. it could be possible for some data to lie outside
> this range (for a forecast dataset it would be impossible to define
> the whole dataset's min and max in advance unless it were defined to
> be much wider than is really useful for visualisation). I think that
> all viz tools should highlight data that are "out of range": it seems
> to me to be a serious flaw to do otherwise as it hides potentially
> vital data.
>
> If a file specifies neither pair of attributes, perhaps there could be
> a community dictionary of suggested ranges per standard name, but this
> is flawed for reasons discussed on this thread. Might be better than
> nothing though.
>
> Regards, Jon
>
> On 1/4/07, David Stuebe <dstuebe at umassd.edu> wrote:
> > Re: attributes for min/max data values for visualization
> > (Chris Webster)
> >
> >
> > This is a very important point that Chris has made regarding the need for
> > min/max data in visualization. For my work with visualization of FVCOM
> > unstructured data, I have only encountered this as an issue while working
> > with multi-domain data sets. Since I have multiple files, one for each
> > domain plus a master file, I only store min/max data in the master file.
> >
> > I have found that the min/max data are only useful in this context, where
> > certain data may not be useful based on its range and can therefore be
> > neglected without reading from disk. My highly >>non<< standard file
> > structure for multi-domain data is to break apart a single file into
> > seperate files which are identical to the original and complete for
> > visualization of that subdomain plus its ghost zones. The master file then
> > contains the data reguarding which cells are ghost cells and the min and max
> > (spatial and data extents) for each sub-domain. This information can be
> > checked first when the user requests a particular plot from the
> > visualization program.
> >
> > Before we get into details like naming schemes for min max values, what are
> > the contexts in which this optional min/max data are useful? This will be an
> > important determining factor in how it is stored.
> >
> > David
> >
> >
> >
> > _______________________________________________
> > CF-metadata mailing list
> > CF-metadata at cgd.ucar.edu
> > http://www.cgd.ucar.edu/mailman/listinfo/cf-metadata
> >
> >
> >
>
>
--
Dr Julian Hill Marine Data Research Scientist
Met Office Hadley Centre for Climate Prediction and Research
FitzRoy Road Exeter EX1 3PB United Kingdom
Tel: +44 (0)1392 884278 Fax: +44(0)1392 885681
Datasets are available from http://www.hadobs.org
Received on Mon Jan 08 2007 - 04:15:34 GMT
