joinExisting is a "low level" aggregation, with little or no smarts. So
it requires that the files be uniform. One can correct some problems,
but only with a lot of work and expertise.
GRIB files can now be handled quite well using the "GRIB feature
collection" feature in version 4.3. This works best from a TDS server, see:
On 8/31/2012 11:09 PM, Paul Graham wrote:
I'm trying to aggregate a model run using NCML with "joinExisting"
where each file contains a single model tau. The filenames correspond
to the tau. Eg. "ec_000.grb, ec_003.grb, ec_006.grb, ec_009.grb ..."
representing tau 0, tau 3, tau 6, tau 9 etc...
I encountered a problem where for one model there are missing files.
Eg. "ec_000.grb, ec_006.grb, ec_009.grb...." (ec_003.grb being the
It appears NCML is unable to handle this situation and generates a
netcdf having a timestep of 6 hours instead of the required 3 hours.
I tried specifying the tau values explicity in the time coordinate
definition using the <values> tag, ie. "<values>0 6 9 12 ....." but
this did not work, while using "<values start="0" increment="3" />"
generated a netcdf with the correct timestep but incorrect times as
the NCML interpreter did not recognise the missing files.
Is there any solution to the problem of aggregating a model run where
there are missing files or a varying timestep (as a result of missing
Any help is appreciated.
netcdf-java mailing list
For list information or to unsubscribe, visit: