netCDF Operators NCO version 5.3.1 howl at the Moon
http://nco.sf.net (Homepage, Mailing lists, Help)
http://github.com/nco/nco (Source Code, Issues, Releases)
What's new?
Version 5.3.1 adds features and fixes to ncremap and ncclimo.
The new autoconversion feature allows these scripts to automagically
convert netCDF4 string variables to netCDF3 character arrays.
The improved treatment of fixed time dimensions allows ncclimo to work
with raw MERRA2 files. Together these improvements make possible
regridding and vertical interpolation of the new formatting of
Copernicus Data Store (CDS) files used for, e.g., ERA5.
Skip this release if these changes do not interest you.
Work on 5.3.3 is underway and includes improved NC_STRING
autoconversion support for other NCO operators.
Enjoy,
Charlie
NEW FEATURES (full details always in ChangeLog):
A. Autoconversion of NC_STRING variables, i.e., automatic conversion
from type NC_STRING to NC_CHAR as dictated by choosing a netCDF3
output from a netCDF4 input file, work for all attributes, but not for
all variables. This is because attributes are at most one-dimensional
and may be of any size whereas variables require gridded dimensions
that usually do not fit the ragged sizes of text strings.
Hence scalar NC_STRING attributes are correctly converted to and
stored as NC_CHAR attributes in the netCDF3 output file, but
not all NC_STRING variables are correctly converted.
As of version 5.3.1, NCO has two distinct levels of support for
autoconversion of NC_STRING variables. ncremap fully supports
autoconversion of scalar NC_STRING variables in both horizontal
regridder and vertical interpolation mode. This means that ncremap
will produce netCDF3 output files from netCDF4 input files (if
requested) and that any NC_STRING variables in the input files will be
translated to NC_CHAR arrays in the netCDF3 output files. There is no
loss of information since the entire contents of the string is
preserved in the character array. Autoconversion stores the character
array with a new dimension named sng_lng_X where X is the smallest
power of ten between 10 and 10000 that is capable of containing the
string contents (including the terminating NUL character).
This type of variable autoconversion fails if the input string length
exceeds 10000 (this is a safety measure, and could be expanded if
people lobby me).
All operators besides ncremap currently retain only the first
character of the input NC_STRING array. This loses information.
We are working to implement the same feature that ncremap enjoys.
Both types of autoconversion fail if the input string variable is
multidimensional, i.e., an array of type NC_STRING. A feature to solve
that case is tractable, though we have not yet received any requests
for it.
Thanks to Waler Hannah (LLNL) for prompting this feature.
http://nco.sf.net/nco.html#autoconversion
http://nco.sf.net/nco.html#ncclimo
B. ncremap's vertical interpolator now automatically extrapolates
variables named "t" (as in ERA5 raw data) as if they were temperature.
Futhermore the interpolator now automatically extrapolates variables
named "H" (as in MERRA2 raw data) and VerticalLayerMidpoint (SCREAM)
as if they were geopotential height. See the documentation for
specifics about the temperature and geopotential height extrapolation
algorithms.
http://nco.sf.net/nco.html#vrt
C. For many years, users at select supercomputer centers have been
able to access the latest versions of NCO scripts directly from
C. Zender's build directories. However, this feature was not
documented until now. Zender's build directories usually contain the
latest NCO snapshot, and so may be unstable and are not for everyday
use. This feature is instead intended to allow users to test and
provide feedback on new features before they are distributed in a
public release.
Invoking ncclimo and ncremap with the --npo (or long-option equivalent
--nco_path_override) flag as the first option (i.e., as $1) causes
these scripts to utilize the binary NCO executables in Zender's build
directories on the following machines at the national supercomputer
centers in parentheses:
acme1 (LLNL), andes (ORNL), chrysalis (ANL), compy (PNNL), derecho
(NCAR), frontier (ORNL), and perlmutter (NERSC).
If the latest features that you want are actually implemented in the
NCO binaries or library, then it may suffice to invoke your own
version of the scripts, as in the first examples below. However, if
the latest features that you want are implemented in the scripts
themselves, then you can either invoke Zender's scripts directly (as
in the second set of examples), or copy those scripts into your own
executable path:
@verbatim
# Access Zender's latest binaries from your default scripts
ncremap --npo -m map.nc in.nc out.nc
ncclimo --npo -P elm -c ${caseid} ...
# Access Zender's latest binaries from Zender's latest scripts
~zender/bin/ncremap --npo -m map.nc in.nc out.nc
~zender/bin/ncclimo --npo -P elm -c ${caseid} ...
@end verbatim
@end example
In both cases, the --npo flag must be the first option invoked.
http://nco.sf.net/nco.html#npo
http://nco.sf.net/nco.html#nco_path_override
BUG FIXES:
A. ncremap's vertical interpolator now understands that the time
dimension in input surface pressure fields may be "fixed".
Previously the interpolator assumed that the temporal dimension, if
any, was an unlimited/record dimension. This could cause it to
miscalculate the number of horizontal spatial dimensions, and then
to fail. Now the interpolator considers any fixed dimension named
"time", or most unlimited dimensions (except certain corner cases) to
be temporal dimension. This allows it to interpolate raw MERRA2 or
ERA5 data. Thanks to Naser Mahfouz (PNNL) for prompting this
fix/feature.
http://nco.sf.net/nco.html#vrt
Full release statement at http://nco.sf.net/ANNOUNCE
--
Charlie Zender, Dept. of Earth System Science
University of California, Irvine 949-891-2429
