NCO User's Guide

Foreword
Summary
1. Introduction
2. Operator Strategies
3. Features common to most operators
4. Reference manual for all operators
5. Contributing
General Index

 -- The Detailed Node Listing ---

Introduction

1.1 Availability
1.2 Operating systems compatible with NCO
1.3 Libraries
1.4 netCDF 2.x vs. 3.x
1.5 Help and Bug reports

Operating systems compatible with NCO

1.2.1 Compiling NCO for Microsoft Windows OS

Operator Strategies

2.1 NCO operator philosophy
2.2 Climate model paradigm
2.3 Temporary output files
2.4 Appending variables to a file
2.5 Addition Subtraction Multiplication and Interpolation
2.6 Averagers vs. Concatenators
2.7 Working with large numbers of input files
2.8 Working with large files
2.9 Approximate NCO memory requirements
2.10 Performance limitations of the operators

Averagers vs. Concatenators

2.6.1 Concatenators ncrcat and ncecat
2.6.2 Averagers ncea, ncra, and ncwa
2.6.3 Interpolator ncflint

Features common to most operators

3.1 Specifying input files
3.2 Accessing files stored remotely
3.3 Retention of remotely retrieved files
3.4 Including/Excluding specific variables
3.5 Including/Excluding coordinate variables
3.6 C & Fortran index conventions
3.7 Hyperslabs
3.8 Wrapped coordinates
3.9 Stride
3.10 Missing values
3.11 Operation Types
3.12 Type conversion
3.13 Suppressing interactive prompts
3.14 History attribute
3.15 NCAR CSM Conventions
3.16 ARM Conventions
3.17 Operator version

Accessing files stored remotely

3.2.1 DODS

Reference manual for all operators

4.1 ncatted netCDF Attribute Editor
4.2 ncdiff netCDF Differencer
4.3 ncea netCDF Ensemble Averager
4.4 ncecat netCDF Ensemble Concatenator
4.5 ncflint netCDF File Interpolator
4.6 ncks netCDF Kitchen Sink
4.7 ncra netCDF Record Averager
4.8 ncrcat netCDF Record Concatenator
4.9 ncrename netCDF Renamer
4.10 ncwa netCDF Weighted Averager

ncwa netCDF Weighted Averager

Masking condition
Normalization

Foreword

Summary

1. Introduction

1.1 Availability
1.2 Operating systems compatible with NCO
1.3 Libraries
1.4 netCDF 2.x vs. 3.x
1.5 Help and Bug reports

1.1 Availability

The documentation for NCO is called the NCO User's Guide. The User's Guide is available in Postscript, HTML, DVI, TeXinfo, and Info formats. These formats are included in the source distribution in the files `nco.ps', `nco.html', `nco.dvi', `nco.texi', and `nco.info*', respectively. All the documentation descends from a single source file, `nco.texi' (1). Hence the documentation in every format is very similar. However, some of the complex mathematical expressions needed to describe ncwa can only be displayed in the Postscript and DVI formats.

If you want to quickly see what the latest improvements in NCO are (without downloading the entire source distribution), visit the NCO homepage at http://nco.sourceforge.net. The HTML version of the User's Guide is also available online through the World Wide Web at URL http://nco.sourceforge.net/nco.html. To build and use NCO, you must have netCDF installed. The netCDF homepage is http://www.unidata.ucar.edu/packages/netcdf.

New NCO releases are announced on the netCDF list and on the nco-announce mailing list http://lists.sourceforge.net/mailman/listinfo/nco-announce.

1.2 Operating systems compatible with NCO

The major prerequisite for installing NCO on a particular platform is the successful, prior installation of the netCDF libraries themselves. Unidata has shown a commitment to maintaining netCDF on all popular UNIX platforms, and is moving towards full support for the Microsoft Windows operating system (OS). Given this, the only difficulty in implementing NCO on a particular platform is standardization of various C and Fortran interface and system calls. The C-code has been tested for ANSI compliance by compiling with GNU gcc -ansi -pedantic. Certain branches in the code were required to satisfy the native SGI and SunOS cc compilers, which are strictly ANSI compliant and do not allow variable-size arrays, a nice feature supported by GNU, UNICOS, Solaris, and AIX compilers.

The most time-intensive portion of NCO execution is spent in arithmetic operations, e.g., multiplication, averaging, subtraction. Until August, 1999, these operations were performed in Fortran by default. This was a design decision based on the speed of Fortran-based object code vs. C-based object code in late 1994. Since 1994 native C compilers have improved their vectorization capabilities and it has become advantageous to replace all Fortran subroutines with C subroutines. Furthermore, this greatly simplifies the task of compiling on nominally unsupported platforms. As of August 1999, NCO is built entirely in C by default. This allows NCO to compile on any machine with an ANSI C compiler. Furthermore, NCO automatically takes advantage of extensions to ANSI C when compiled with the GNU compiler collection, GCC.

As of July 2000 and NCO version 1.2, NCO no longer supports performing arithmetic operations in Fortran to improve speed. Supporting Fortran involves maintaining two sets of routines for every arithmetic operation. The USE_FORTRAN_ARITHMETIC flag is retained in the Makefile and the file containing the Fortran code, `nc_fortran.F', is still distributed with NCO in case a volunteer decides to resurrect them. If you would like to volunteer to maintain `nc_fortran.F' please contact me. Otherwise the Fortran hooks will be completely removed in the next major release.

1.2.1 Compiling NCO for Microsoft Windows OS

1.2.1 Compiling NCO for Microsoft Windows OS

NCO has been successfully ported and tested on the Microsoft Windows NT 4.0 operating system. The switches necessary to accomplish this are included in the standard distribution of NCO. Using the freely available Cygwin (formerly gnu-win32) development environment (2), the compilation process is very similar to installing NCO on a UNIX system. The preprocessor token PVM_ARCH should be set to WIN32. Note that defining WIN32 has the side effect of disabling Internet features of NCO (see below). Unless you have a Fortran compiler (like g77 or f90) available, no other tokens are required. Users with fast Fortran compilers may wish to activate the Fortran arithmetic routines. To do this, define the preprocessor token USE_FORTRAN_ARITHMETIC in the makefile which comes with NCO, `Makefile', or in the compilation shell.

The least portable section of the code is the use of standard UNIX and Internet protocols (e.g., ftp, rcp, scp, getuid, gethostname, and header files `<arpa/nameser.h>' and `<resolv.h>'). Fortunately, these UNIXy calls are only invoked by the single NCO subroutine which is responsible for retrieving files stored on remote systems (see section 3.2 Accessing files stored remotely). In order to support NCO on the Microsoft Windows platforms, this single feature was disabled (on Windows OS only). This was required by Cygwin 18.x--newer versions of Cygwin may support these protocols (let me know if this is the case). The NCO operators should behave identically on Windows and UNIX platforms in all other respects.

1.3 Libraries

1.4 netCDF 2.x vs. 3.x

However, the ability to compile NCO with only netCDF 2.x calls is worth maintaining because HDF version 4 (4) (available from HDF) supports only the netCDF 2.x library calls (see http://hdf.ncsa.uiuc.edu/UG41r3_html/SDS_SD.fm12.html#47784). Note that there are multiple versions of HDF. Currently HDF version 4.x supports netCDF 2.x and thus NCO version 1.2.x. If NCO version 1.2.x (or earlier) is built with only netCDF 2.x calls then all NCO operators should work with HDF4 files as well as netCDF files (5). The preprocessor token NETCDF2_ONLY exists in NCO version 1.2.x to eliminate all netCDF 3.x calls. Only versions of NCO numbered 1.2.x and earlier have this capability. The NCO 1.2.x branch will be maintained with bugfixes only (no new features) until HDF begins to fully support the netCDF 3.x interface (which is employed by NCO 2.x). If, at compilation time, NETCDF2_ONLY is defined, then NCO version 1.2.x will not use any netCDF 3.x calls and, if linked properly, the resulting NCO operators will work with HDF4 files. The `Makefile' supplied with NCO 1.2.x has been written to simplify building in this HDF capability. When NCO is built with make HDF4=Y, the `Makefile' will set all required preprocessor flags and library links to build with the HDF4 libraries (which are assumed to reside under /usr/local/hdf4, edit the `Makefile' to suit your installation).

HDF version 5.x became available in 1999, but did not support netCDF (or, for that matter, Fortran) as of December 1999. By early 2001, HDF version 5.x did support Fortran90. However, support for netCDF 3.x in HDF 5.x is incomplete. Much of the HDF5-netCDF3 interface is complete, however, and it may be separately downloaded from the HDF5-netCDF website. Now that NCO uses only netCDF 3.x system calls we are eager for HDF5 to complete their netCDF 3.x support.

1.5 Help and Bug reports

If you would like NCO to include a new feature, first check to see if that feature is already on the TODO list. If it is, please consider implementing that feature yourself and sending us the patch! If the feature is not yet on the list then send a note to the NCO Discussion forum.

Please read the manual before reporting a bug or posting a request for help. Sending questions whose answers are not in the manual is the best way to motivate us to write more documentation. We would also like to accentuate the contrapositive of this statement. If you think you have found a real bug the most helpful thing you can do is simplify the problem to a manageable size and report it. The first thing to do is to make sure you are running the latest publicly released version of NCO.

Once you have read the manual, if you are still unable to get NCO to perform a documented function, write help request. Follow the same procedure as described below for reporting bugs (after all, it might be a bug). That is, describe what you are trying to do, and inlude the complete commands (with `-D 5'), error messages, and version of NCO. Post your help request to the NCO Help forum.

If you think you are using the right command, but NCO is misbehaving, then you might have found a bug. A core dump, sementation violation, or incorrect numerical answers is always considered a high priority bug. How do you simplify a problem that may be revealing a bug? Cut out extraneous variables, dimensions, and metadata from the offending files and re-run the command until it no longer breaks. Then back up one step and report the problem. Usually the file(s) will be very small, i.e., one variable with one or two small dimensions ought to suffice. Include in the report your run-time environment, the exact error messages (and run the operator with `-D 5' to increase the verbosity of the debugging output), and a copy, or the publically accessible location, of the file(s). Post the bug report to the NCO Project buglist.

2. Operator Strategies

2.1 NCO operator philosophy
2.2 Climate model paradigm
2.3 Temporary output files
2.4 Appending variables to a file
2.5 Addition Subtraction Multiplication and Interpolation
2.6 Averagers vs. Concatenators
2.7 Working with large numbers of input files
2.8 Working with large files
2.9 Approximate NCO memory requirements
2.10 Performance limitations of the operators

2.1 NCO operator philosophy

The main design goal has been to produce operators that can be invoked from the command line to perform useful operations on netCDF files. Many scientists work with models and observations which produce too much data to analyze in tabular format. Thus, it is often natural to reduce and massage this raw or primary level data into summary, or second level data, e.g., temporal or spatial averages. These second level data may become the inputs to graphical and statistical packages, and are often more suitable for archival and dissemination to the scientific community. NCO performs a suite of operations useful in manipulating data from the primary to the second level state. Higher level interpretive languages (e.g., IDL, Yorick, Matlab, NCL, Perl, Python), and lower level compiled languages (e.g., C, Fortran) can always perform any task performed by NCO, but often with more overhead. NCO, on the other hand, is limited to a much smaller set of arithmetic and metadata operations than these full blown languages.

Another goal has been to implement enough command line switches so that frequently used sequences of these operators can be executed from a shell script or batch file. Finally, NCO was written to consume the absolute minimum amount of system memory required to perform a given job. The arithmetic operators are extremely efficient; their exact memory usage is detailed in 2.9 Approximate NCO memory requirements.

2.2 Climate model paradigm

NCO was developed at NCAR to aid analysis and manipulation of datasets produced by General Circulation Models (GCMs). Datasets produced by GCMs share many features with all gridded scientific datasets and so provide a useful paradigm for the explication of the NCO operator set. Examples in this manual use a GCM paradigm because latitude, longitude, time, temperature and other fields related to our natural environment are as easy to visualize for the layman as the expert.

2.3 Temporary output files

Other safeguards exist to protect the user from inadvertently overwriting data. If the output-file specified for a command is a pre-existing file, then the operator will prompt the user whether to overwrite (erase) the existing output-file, attempt to append to it, or abort the operation. However, in processing large amounts of data, too many interactive questions can be a curse to productivity. Therefore NCO also implements two ways to override its own safety features, the `-O' and `-A' switches. Specifying `-O' tells the operator to overwrite any existing output-file without prompting the user interactively. Specifying `-A' tells the operator to attempt to append to any existing output-file without prompting the user interactively. These switches are useful in batch environments because they suppress interactive keyboard input.

2.4 Appending variables to a file

2.5 Addition Subtraction Multiplication and Interpolation

Users comfortable with NCO semantics may find it easier to perform some simple mathematical operations in NCO rather than higher level languages. ncdiff (see section 4.2 ncdiff netCDF Differencer) can be used for subtraction and broadcasting. ncflint (see section 4.5 ncflint netCDF File Interpolator) can be used for addition, subtraction, multiplication and interpolation. Sequences of these commands can accomplish simple but powerful operations at the command line.

2.6 Averagers vs. Concatenators

The most frequently used operators of NCO are probably the averagers and concatenators. Because there are so many permutations of averaging (e.g., across files, within a file, over the record dimension, over other dimensions, with or without weights and masks) and of concatenating (across files, along the record dimension, along other dimensions), there are currently no fewer than five operators which tackle these two purposes: ncra, ncea, ncwa, ncrcat, and ncecat. These operators do share many capabilities (9), but each has its unique specialty. Two of these operators, ncrcat and ncecat, are for concatenating hyperslabs across files. The other two operators, ncra and ncea, are for averaging hyperslabs across files (10). First, let's describe the concatenators, then the averagers.

2.6.1 Concatenators ncrcat and ncecat
2.6.2 Averagers ncea, ncra, and ncwa
2.6.3 Interpolator ncflint

2.6.1 Concatenators ncrcat and ncecat

Joining independent files together along a record coordinate is called concatenation. ncrcat is designed for concatenating record variables, while ncecat is designed for concatenating fixed length variables. Consider 5 files, `85.nc', `86.nc', ... `89.nc' each containing a year's worth of data. Say you wish to create from them a single file, `8589.nc' containing all the data, i.e., spanning all 5 years. If the annual files make use of the same record variable, then ncrcat will do the job nicely with, e.g., ncrcat 8?.nc 8589.nc. The number of records in the input files is arbitrary and can vary from file to file. See section 4.8 ncrcat netCDF Record Concatenator, for a complete description of ncrcat.

However, suppose the annual files have no record variable, and thus their data is all fixed length. For example, the files may not be conceptually sequential, but rather members of the same group, or ensemble. Members of an ensemble may have no reason to contain a record dimension. ncecat will create a new record dimension (named record by default) with which to glue together the individual files into the single ensemble file. If ncecat is used on files which contain an existing record dimension, that record dimension will be converted into a fixed length dimension of the same name and a new record dimension will be created. Consider five realizations, `85a.nc', `85b.nc', ... `85e.nc' of 1985 predictions from the same climate model. Then ncecat 85?.nc 85_ens.nc glues the individual realizations together into the single file, `85_ens.nc'. If an input variable was dimensioned [lat,lon], it will have dimensions [record,lat,lon] in the output file. A restriction of ncecat is that the hyperslabs of the processed variables must be the same from file to file. Normally this means all the input files are the same size, and contain data on different realizations of the same variables. See section 4.4 ncecat netCDF Ensemble Concatenator, for a complete description of ncecat.

Note that ncrcat cannot concatenate fixed-length variables, whereas ncecat can concatenate both fixed-length and record variables. To conserve system memory, use ncrcat rather than ncecat when concatenating record variables.

2.6.2 Averagers ncea, ncra, and ncwa

The differences between the averagers ncra and ncea are analogous to the differences between the concatenators. ncra is designed for averaging record variables from at least one file, while ncea is designed for averaging fixed length variables from multiple files. ncra performs a simple arithmetic average over the record dimension of all the input files, with each record having an equal weight in the average. ncea performs a simple arithmetic average of all the input files, with each file having an equal weight in the average. Note that ncra cannot average fixed-length variables, but ncea can average both fixed-length and record variables. To conserve system memory, use ncra rather than ncea where possible (e.g., if each input-file is one record long). The file output from ncea will have the same dimensions (meaning dimension names as well as sizes) as the input hyperslabs (see section 4.3 ncea netCDF Ensemble Averager, for a complete description of ncea). The file output from ncra will have the same dimensions as the input hyperslabs except for the record dimension, which will have a size of 1 (see section 4.7 ncra netCDF Record Averager, for a complete description of ncra).

2.6.3 Interpolator ncflint

ncflint can interpolate data between or two files. Since no other operators have this ability, the description of interpolation is given fully on the ncflint reference page (see section 4.5 ncflint netCDF File Interpolator). Note that this capability also allows ncflint to linearly rescale any data in a netCDF file, e.g., to convert between differing units.

2.7 Working with large numbers of input files

Occasionally one desires to digest (i.e., concatenate or average) hundreds or thousands of input files. One brave user, for example, recently created a five year time-series of satellite observations by using ncecat to join thousands of daily data files together. Unfotunately, data archives (e.g., NASA EOSDIS) are unlikely to distribute netCDF files conveniently named in a format the `-n loop' switch (which automatically generates arbitrary numbers of input filenames) understands. If there is not a simple, arithmetic pattern to the input filenames (e.g., `h00001.nc', `h00002.nc', ... `h90210.nc') then the `-n loop' switch is useless. Moreover, when the input files are so numerous that the input filenames are too lengthy (when strung together as a single argument) to be passed by the calling shell to the NCO operator (11), then the following strategy has proven useful to specify the input filenames to NCO. Write a script that creates symbolic links between the irregular input filenames and a set of regular, arithmetic filenames that `-n loop' switch understands. The NCO operator will then succeed at automatically generating the filnames with the `-n loop' option (which circumvents any OS and shell limits on command line size). You can remove the symbolic links once the operator completes its task.

2.8 Working with large files

Large files are those files that are comparable in size to the amount of memory (RAM) in your computer. Many users of NCO work with files larger than 100 Mb. Files this large not only push the current edge of storage technology, they present special problems for programs which attempt to access the entire file at once, such as ncea, and ncecat. If you need to work with a 300 Mb file on a machine with only 32 Mb of memory then you will need large amounts of swap space (virtual memory on disk) and NCO will work slowly, or else NCO will fail. There is no easy solution for this and the best strategy is to work on a machine with massive amounts of memory and swap space. That is, if your local machine has problems working with large files, try running NCO from a more powerful machine, such as a network server. Certain machine architectures, e.g., Cray UNICOS, have special commands which allow one to increase the amount of interactive memory. If you get a core dump on a Cray system (e.g., `Error exit (core dumped)'), try increasing the available memory by using the ilimit command.

The speed of the NCO operators also depends on file size. When processing large files the operators may appear to hang, or do nothing, for large periods of time. In order to see what the operator is actually doing, it is useful to activate a more verbose output mode. This is accomplished by supplying a number greater than 0 to the `-D debug_level' switch. When the debug_level is nonzero, the operators report their current status to the terminal through the stderr facility. Using `-D' does not slow the operators down. Choose a debug_level between 1 and 3 for most situations, e.g., ncea -D 2 85.nc 86.nc 8586.nc. A full description of how to estimate the actual amount of memory the multi-file NCO operators consume is given in 2.9 Approximate NCO memory requirements.

2.9 Approximate NCO memory requirements

The multi-file operators currently comprise the record operators, ncra and ncrcat, and the ensemble operators, ncea and ncecat, The record operators require much less memory than the ensemble operators. This is because the record operators are designed to operate on a single record of a file at a time, while the ensemble operators must retrieve an entire variable at a time into memory. Let MS be the peak sustained memory demand of an operator, FT be the memory required to store the entire contents of all the variables to be processed in an input file, FR be the memory required to store the entire contents of a single record of each of the variables to be processed in an input file, VR be the memory required to store a single record of the largest record variable to be processed in an input file, VT be the memory required to store the largest variable to be processed in an input file, VI be the memory required to store the largest variable which is not processed, but is copied from the initial file to the output file. All operators require MI = VI during the initial copying of variables from the first input file to the output file. This is the initial (and transient) memory demand. The sustained memory demand is that memory required by the operators during the processing (i.e., averaging, concatenation) phase which lasts until all the input files have been processed. The operators have the following memory requirements: ncrcat requires MS <= VR. ncecat requires MS <= VT. ncra requires MS = 2FR + VR. ncea requires MS = 2FT + VT. ncdiff requires MS <= 2VT. ncflint requires MS <= 2VT. Note that only variables which are processed, i.e., averaged or concatenated, contribute to MS. Memory is never allocated to hold variables which do not appear in the output file (see section 3.4 Including/Excluding specific variables).

2.10 Performance limitations of the operators

1. No buffering of data is performed during ncvarget and ncvarput operations. Hyperslabs too large too hold in core memory will suffer substantial performance penalties because of this.

2. Since coordinate variables are assumed to be monotonic, the search for bracketing the user-specified limits should employ a quicker algorithm, like bisection, than the two-sided incremental search currently implemented.

3. C_format, FORTRAN_format, signedness, scale_format and add_offset attributes are ignored by ncks when printing variables to screen.

4. Some random access operations on large files on certain architectures (e.g., 400 Mb on UNICOS) are much slower with these operators than with similar operations performed using languages that bypass the netCDF interface (e.g., Yorick). The cause for this is not understood at present.

3. Features common to most operators

Many features have been implemented in more than one operator and are described here for brevity. The description of each feature is preceded by a box listing the operators for which the feature is implemented. Command line switches for a given feature are consistent across all operators wherever possible. If no "key switches" are listed for a feature, then that particular feature is automatic and cannot be controlled by the user.

3.1 Specifying input files
3.2 Accessing files stored remotely
3.3 Retention of remotely retrieved files
3.4 Including/Excluding specific variables
3.5 Including/Excluding coordinate variables
3.6 C & Fortran index conventions
3.7 Hyperslabs
3.8 Wrapped coordinates
3.9 Stride
3.10 Missing values
3.11 Operation Types
3.12 Type conversion
3.13 Suppressing interactive prompts
3.14 History attribute
3.15 NCAR CSM Conventions
3.16 ARM Conventions
3.17 Operator version

3.1 Specifying input files

ncra 85.nc 86.nc 87.nc 88.nc 89.nc 8589.nc
ncra 8[56789].nc 8589.nc
ncra -p input-path 85.nc 86.nc 87.nc 88.nc 89.nc 8589.nc
ncra -n 5,2,1 85.nc 8589.nc

The last method passes (with `-n') syntax concisely describing the entire set of filenames (12). This option is only available with the multi-file operators: ncra, ncrcat, ncea, and ncecat. By definition, multi-file operators are able to process an arbitrary number of input-files. This option is very useful for abbreviating lists of filenames representable as alphanumeric_prefix+numeric_suffix+`.'+filetype where alphanumeric_prefix is a string of arbitrary length and composition, numeric_suffix is a fixed width field of digits, and filetype is a standard filetype indicator. For example, in the file `ccm3_h0001.nc', we have alphanumeric_prefix = `ccm3_h', numeric_suffix = `0001', and filetype = `nc'.

NCO is able to decode lists of such filenames encoded using the `-n' option. The simpler (3-argument) `-n' usage takes the form -n file_number,digit_number,numeric_increment where file_number is the number of files, digit_number is the fixed number of numeric digits comprising the numeric_suffix, and numeric_increment is the constant, integer-valued difference between the numeric_suffix of any two consecutive files. The value of alphanumeric_prefix is taken from the input file, which serves as a template for decoding the filenames. In the example above, the encoding -n 5,2,1 along with the input file name `85.nc' tells NCO to construct five (5) filenames identical to the template `85.nc' except that the final two (2) digits are a numeric suffix to be incremented by one (1) for each successive file. Currently filetype may be either be empty, `nc', `cdf', `hdf', or `hd5'. If present, these filetype suffixes (and the preceding `.') are ignored by NCO as it uses the `-n' arguments to locate, evaluate, and compute the numeric_suffix component of filenames.

Recently the `-n' option has been extended to allow convenient specification of filenames with "circular" characteristics. This means it is now possible for NCO to automatically generate filenames which increment regularly until a specified maximum value, and then wrap back to begin again at a specified minimum value. The corresponding `-n' usage becomes more complex, taking one or two additional arguments for a total of four or five, respectively: -n file_number,digit_number,numeric_increment[,numeric_max[,numeric_min]] where numeric_max, if present, is the maximum integer-value of numeric_suffix and numeric_min, if present, is the minimum integer-value of numeric_suffix. Consider, for example, the problem of specifying non-consecutive input files where the filename suffixes end with the month index. In climate modeling it is common to create summertime and wintertime averages which contain the averages of the months June--July--August, and December--January--February, respectively:

ncra -n 3,2,1 85_06.nc 85_0608.nc ncra -n 3,2,1,12 85_12.nc 85_1202.nc ncra -n 3,2,1,12,1 85_12.nc 85_1202.nc

The first example shows that three arguments to the `-n' option suffice to specify consecutive months (06, 07, 08) which do not "wrap" back to a minimum value. The second example shows how to use the optional fourth and fifth elements of the `-n' option to specify a wrap value to NCO. The fourth argument to `-n', if present, specifies the maximum integer value of numeric_suffix. In this case the maximum value is 12, and will be formatted as `12' in the filename string. The fifth argument to `-n', if present, specifies the minimum integer value of numeric_suffix. The default minimum filename suffix is 1, which is formatted as `01' in this case. Thus the second and third examples have the same effect, that is, they automatically generate, in order, the filenames `85_12.nc', `85_01.nc', and `85_02.nc' as input to NCO.

3.2 Accessing files stored remotely

To access a file on NCAR's MSS, specify the full MSS pathname of the remote file. NCO will attempt to detect whether the local machine has direct (synchronous) MSS access. In this case, NCO attempts to use the NCAR msrcp command (13), or, failing that, /usr/local/bin/msread. Otherwise NCO attempts to retrieve the MSS file through the (asynchronous) Masnet Interface Gateway System (MIGS) using the nrnet command.

The following examples show how one might analyze files stored on remote systems.

ncks -H -l ./ ftp://ftp.cgd.ucar.edu/pub/zender/nco/in.nc ncks -H -l ./ dust.ps.uci.edu:/home/zender/nco/in.nc ncks -H -l ./ /ZENDER/nco/in.nc ncks -H -l ./ mss:/ZENDER/nco/in.nc ncks -H -l ./ -p http://www.cdc.noaa.gov/cgi-bin/nph-nc/Datasets/- ncep.reanalysis.dailyavgs/surface air.sig995.1975.nc

The first example will work verbatim on your system if your system is connected to the Internet and is not behind a firewall. The second example will work on your system if you have rcp or scp access to the machine dust.ps.uci.edu. The third example will work from NCAR computers with local access to the msrcp, msread, or nrnet commands. The fourth command will work if your local version of NCO was built with DODS capability (see section 3.2.1 DODS). The above commands can be rewritten using the `-p input-path' option as follows:

ncks -H -p ftp://ftp.cgd.ucar.edu/pub/zender/nc -l ./ in.nc ncks -H -p dust.ps.uci.edu:/home/zender/nc -l ./ in.nc ncks -H -p /ZENDER/nco -l ./ in.nc ncks -H -p mss:/ZENDER/nco -l ./ in.nc

Using `-p' is recommended because it clearly separates the input-path from the filename itself, sometimes called the stub. When input-path is not explicitly specified using `-p', NCO internally generates an input-path from the first input filename. The automatically generated input-path is constructed by stripping the input filename of everything following the final `/' character (i.e., removing the stub). The `-l output-path' option tells NCO where to store the remotely retrieved file and the output file. Often the path to a remotely retrieved file is quite different than the path on the local machine where you would like to store the file. If `-l' is not specified then NCO internally generates an output-path by simply setting output-path equal to input-path stripped of any machine names. If `-l' is not specified and the remote file resides on the NCAR MSS system, then the leading character of input-path, `/', is also stripped from output-path. Specifying output-path as `-l ./' tells NCO to store the remotely retrieved file and the output file in the current directory. Note that `-l .' is equivalent to `-l ./' though the latter is syntactically more clear.

3.2.1 DODS

3.2.1 DODS

Once NCO is DODS-enabled the operators are DODS clients. All DODS clients have network transparent access to any files controlled by a DODS server. Simply specify the path to the file in URL notation

ncks -C -d lon,0 -v lon -l ./ -p http://www.cdc.noaa.gov/cgi-bin/nph-nc/ Datasets/ncep.reanalysis.dailyavgs/surface air.sig995.1975.nc foo.nc

NCO operates on these remote files without having to transfer the files to the local disk. DODS causes all the I/O to appear to NCO as if the files were local. The advantage to this is that only the required data (e.g., the variable or hyperslab specified) are transferred over the network. The advantages of this are obvious if you are examining small parts of large files stored at remote locations.

Note that the remote retrieval features of NCO can be used to retrieve any file, including non-netCDF files, via SSH, anonymous FTP, or msrcp. Often this method is quicker than using a browser, or running an FTP session from a shell window yourself. For example, say you want to obtain a JPEG file from a weather server.

ncks -p ftp://weather.edu/pub/pix/jpeg -l ./ storm.jpg

In this example, ncks automatically performs an anonymous FTP login to the remote machine and retrieves the specified file. When ncks attempts to read the local copy of `storm.nc' as a netCDF file, it fails and exits, leaving `storm.nc' in the current directory.

3.3 Retention of remotely retrieved files

3.4 Including/Excluding specific variables

3.5 Including/Excluding coordinate variables

3.6 C & Fortran index conventions

ncra -d time,5,7 85.nc 85_JJA.nc
ncra -F -d time,6,8 85.nc 85_JJA.nc

Printing variable three_dmn_var in file `in.nc' first with C indexing conventions, then with Fortran indexing conventions results in the following output formats:

% ncks -H -v three_dmn_var in.nc % lat[0]=-90 lev[0]=1000 lon[0]=-180 three_dmn_var[0]=0 ... % ncks -F -H -v three_dmn_var in.nc % lon(1)=-180 lev(1)=1000 lat(1)=-90 three_dmn_var(1)=0

3.7 Hyperslabs

Coordinate values should be specified using real notation with a decimal point required in the value, whereas dimension indices are specified using integer notation without a decimal point. Note that this convention is only to differentiate coordinate values from dimension indices, and is independent of the actual type of netCDF coordinate variables, if any. For a given dimension, the specified limits must both be coordinate values (with decimal points) or dimension indices (no decimal points).

User-specified coordinate limits are promoted to double precision values while searching for the indices which bracket the range. Thus, hyperslabs on coordinates of type NC_BYTE and NC_CHAR are computed numerically rather than lexically, so the results are unpredictable.

The relative magnitude of min and max indicate to the operator whether to expect a wrapped coordinate (see section 3.8 Wrapped coordinates), such as longitude. If min > max, the NCO expects the coordinate to be wrapped, and a warning message will be printed. When this occurs, NCO selects all values outside the domain [max < min], i.e., all the values exclusive of the values which would have been selected if min and max were swapped. If this seems confusing, test your command on just the coordinate variables with ncks, and then examine the output to ensure NCO selected the hyperslab you expected (coordinate wrapping is only supported by ncks).

Because of the way wrapped coordinates are interpreted, it is very important to make sure you always specify hyperslabs in the monotonically increasing sense, i.e., min < max (even if the underlying coordinate variable is monotonically decreasing). The only exception to this is when you are indeed specifying a wrapped coordinate. The distinction is crucial to understand because the points selected by, e.g., -d longitude,50.,340., are exactly the complement of the points selected by -d longitude,340.,50..

Not specifying any hyperslab option is equivalent to specifying full ranges of all dimensions. This option may be specified more than once in a single command (each hyperslabed dimension requires its own -d option).

3.8 Wrapped coordinates

Assume the domain of the monotonically increasing longitude coordinate lon is 0 < lon < 360. ncks will extract a hyperslab which crosses the Greenwich meridian simply by specifying the westernmost longitude as min and the easternmost longitude as max. Thus, the following commands extract a hyperslab containing the Saharan desert:

ncks -d lon,340.,50. in.nc out.nc ncks -d lon,340.,50. -d lat,10.,35. in.nc out.nc

The first example selects data in the same longitude range as the Sahara. The second example further constrains the data to having the same latitude as the Sahara. The coordinate lon in the output-file, `out.nc', will no longer be monotonic! The values of lon will be, e.g., `340, 350, 0, 10, 20, 30, 40, 50'. This can have serious implications should you run `out.nc' through another operation which expects the lon coordinate to be monotonically increasing. Fortunately, the chances of this happening are slim, since lon has already been hyperslabbed, there should be no reason to hyperslab lon again. Should you need to hyperslab lon again, be sure to give dimensional indices as the hyperslab arguments, rather than coordinate values (see section 3.7 Hyperslabs).

3.9 Stride

The stride is specified as the optional fourth argument to the `-d' hyperslab specification: -d dim,[min][,[max]][,[stride]]. Specify stride as an integer (i.e., no decimal point) following the third comma in the `-d' argument. There is no default value for stride. Thus using `-d time,,,2' is valid but `-d time,,,2.0' and `-d time,,,' are not. When stride is specified but min is not, there is an ambiguity as to whether the extracted hyperslab should begin with (using C-style, 0-based indexes) element 0 or element `stride-1'. NCO must resolve this ambiguity and it chooses element 0 as the first element of the hyperslab when min is not specified. Thus `-d time,,,stride' is syntactically equivalent to `-d time,0,,stride'. This means, for example, that specifying the operation `-d time,,,2' on the array `1,2,3,4,5' selects the hyperslab `1,3,5'. To obtain the hyperslab `2,4' instead, simply explicitly specify the starting index as 1, i.e., `-d time,1,,2'.

For example, consider a file `8501_8912.nc' which contains 60 consecutive months of data. Say you wish to obtain just the March data from this file. Using 0-based subscripts (see section 3.6 C & Fortran index conventions) these data are stored in records 2, 14, ... 50 so the desired stride is 12. Without the stride option, the procedure is very awkward. One could use ncks five times and then use ncrcat to concatenate the resulting files together:

foreach idx (02 14 26 38 50) ncks -d time,$idx 8501_8912.nc foo.$idx end ncrcat foo.?? 8589_03.nc rm foo.??

With the stride option, ncks performs this hyperslab extraction in one operation:

ncks -d time,2,,12 8501_8912.nc 8589_03.nc

See section 4.6 ncks netCDF Kitchen Sink, for more information on ncks.

The stride option is supported by ncra and ncrcat for the record dimension only. This makes it possible, for instance, to average or concatenate regular intervals across multi-file input data sets.

ncra -F -d time,3,,12 85.nc 86.nc 87.nc 88.nc 89.nc 8589_03.nc ncrcat -F -d time,3,,12 85.nc 86.nc 87.nc 88.nc 89.nc 8503_8903.nc

3.10 Missing values

The phrase missing data refers to data points that are missing, invalid, or for any reason not intended to be arithmetically processed in the same fashion as valid data. The NCO arithmetic operators attempt to handle missing data in an intelligent fashion. There are four steps in the NCO treatment of missing data:

1. Identifying variables which may contain missing data.

   NCO follows the convention that missing data should be stored with the missing_value specified in the variable's missing_value attribute (14). The only way NCO recognizes that a variable may contain missing data is if the variable has a missing_value attribute. In this case, any elements of the variable which are numerically equal to the missing_value are treated as missing data.

2. Converting the missing_value to the type of the variable, if neccessary.

   Consider a variable var of type var_type with a missing_value attribute of type att_type containing the value missing_value. As a guideline, the type of the missing_value attribute should be the same as the type of the variable it is attached to. If var_type equals att_type then NCO straightforwardly compares each value of var to missing_value to determine which elements of var are to be treated as missing data. If not, then NCO will internally convert att_type to var_type by using the implicit conversion rules of C, or, if att_type is NC_CHAR (15), by typecasting the results of the C function strtod(missing_value). You may use the NCO operator ncatted to change the missing_value attribute and all data whose data is missing_value to a new value (see section 4.1 ncatted netCDF Attribute Editor).

3. Identifying missing data during arithmetic operations.

   When an NCO arithmetic operator is processing a variable var with a missing_value attribute, it compares each value of var to missing_value before performing an operation. Note the missing_value comparison inflicts a performance penalty on the operator. Arithmetic processing of variables which contain the missing_value attribute always incurs this penalty, even when none of the data is missing. Conversely, arithmetic processing of variables which do not contain the missing_value attribute never incurs this penalty. In other words, do not attach a missing_value attribute to a variable which does not contain missing data. This exhortation can usually be obeyed for model generated data, but it may be harder to know in advance whether all observational data will be valid or not.

4. Treatment of any data identified as missing in arithmetic operators.

   NCO averagers (ncra, ncea, ncwa) do not count any element with the value missing_value towards the average. ncdiff and ncflint define a missing_value result when either of the input values is a missing_value. Sometimes the missing_value may change from file to file in a multi-file operator, e.g., ncra. NCO is written to account for this (it always compares a variable to the missing_value assigned to that variable in the current file). Suffice it to say that, in all known cases, NCO does "the right thing".

3.11 Operation Types

avg

Mean value (default)

sqravg

Square of the mean

avgsqr

Mean of sum of squares

max

Maximium value

min

Minimium value

rms

Root-mean-square (normalized by N)

rmssdn

Root-mean square normalized by N-1

sqrt

Square root of the mean

ttl

Sum of values

<p><b>Note for HTML user's</b>: <br>The definition of mathematical operations involving rank reduction (e.g., averaging) relies heavily on mathematical expressions which cannot be easily represented in HTML. <b>See the printed manual for complete documentation.</b> Note for Info user's: The definition of mathematical operations involving rank reduction (e.g., averaging) relies heavily on mathematical expressions which cannot be easily represented in Info. See the printed manual for complete documentation.

The definitions of some of these operations are not universally useful. Mostly they were chosen to facilitate standard statistical computations within the NCO framework. We are open to redefining and or adding to the above. If you are interested in having other statistical quantities defined in NCO please contact the NCO project (see section 1.5 Help and Bug reports).

EXAMPLES

Suppose you wish to examine the variable prs_sfc(time,lat,lon) which contains a time series of the surface pressure as a function of latitude and longitude. Find the minimium value of prs_sfc over all dimensions:

ncwa -y min -v prs_sfc in.nc foo.nc

Find the maximum value of prs_sfc at each time interval for each latitude:

ncwa -y max -v prs_sfc -a lon in.nc foo.nc

Find the root-mean-square value of the time-series of prs_sfc at every gridpoint:

ncra -y rms -v prs_sfc in.nc foo.nc ncwa -y rms -v prs_sfc -a time in.nc foo.nc

The previous two commands give the same answer but ncra is preferred because it has a smaller memory footprint. Also, ncra leaves the (degenerate) time dimension in the output file (which is usually useful) whereas ncwa removes the time dimension.

These operations work as expected in multi-file operators. Suppose that prs_sfc is stored in multiple timesteps per file across multiple files, say `jan.nc', `feb.nc', `march.nc'. We can now find the three month maximium surface pressure at every point.

ncea -y max -v prs_sfc jan.nc feb.nc march.nc out.nc

It is possible to use a combination of these operations to compute the variance and standard deviation of a field stored in a single file or across multiple files. The procedure to compute the temporal standard deviation of the surface pressure at all points in a single file `in.nc' involves three steps.

ncwa -O -v prs_sfc -a time in.nc out.nc ncdiff -O -v prs_sfc in.nc out.nc out.nc ncra -O -y rmssdn out.nc out.nc

First the output file `out.nc' is contructed containing the temporal mean of prs_sfc. Next `out.nc' is overwritten with the deviation from the mean. Finally `out.nc' is overwritten with the root-mean-square of itself. Note the use of `-y rmssdn' (rather than `-y rms') in the final step. This ensures the standard deviation is correctly normalized by one fewer than the number of time samples. The procedure to compute the variance is identical except for the use of `-y var' instead of `-y rmssdn' in the final step.

The procedure to compute the spatial standard deviation of a field in a single file `in.nc' involves three steps.

ncwa -O -v prs_sfc,gw -a lat,lon -w gw in.nc out.nc ncdiff -O -v prs_sfc,gw in.nc out.nc out.nc ncwa -O -y rmssdn -v prs_sfc -a lat,lon -w gw out.nc out.nc

First the appropriately weighted (with `-w gw') spatial mean values are written to the output file. This example includes the use of a weighted variable specified with `-w gw'. When using weights to compute standard deviations one must remember to include the weights in the initial output files so that they may be used again in the final step. The initial output file is then overwritten with the gridpoint deviations from the spatial mean. Finally the root-mean-square of the appropriately weighted spatial deviations is taken.

The procedure to compute the standard deviation of a time-series across multiple files involves one extra step since all the input must first be collected into one file.

ncrcat -O -v tpt in.nc in.nc foo1.nc ncwa -O -a time foo1.nc foo2.nc ncdiff -O -v tpt foo1.nc foo2.nc foo2.nc ncra -O -y rmssdn foo2.nc out.nc

The first step assembles all the data into a single file. This may require a lot of temporary disk space, but is more or less required by the ncdiff operation in the fourth step.

3.12 Type conversion

Type conversion during arithmetic in the languages C and Fortran is performed only when necessary. All operands in an operation are converted to the most precise type before the operation takes place. However, following this parsimonious conversion rule dogmatically results in numerous headaches. For example, the average of the two NC_SHORTs 17000s and 17000s results in garbage since the intermediate value which holds their sum is also of type NC_SHORT and thus cannot represent values greater than 32,767 (16). There are valid reasons for expecting this operation to succeed and the NCO philosophy is to make operators do what you want, not what is most pure. Thus, unlike C and Fortran, but like many other higher level interpreted languages, NCO arithmetic operators will perform automatic type conversion when all the following conditions are met (17):

1. The operator is ncea, ncra, or ncwa. ncdiff is not included because subtraction does not benefit from type conversion.
2. The arithmetic operation could benefit from type conversion. Operations that could benefit (e.g., from larger representable sums) include averaging, summation, or any "hard" arithmetic. Type conversion does not benefit searching for minima and maxima (`-y min', or `-y max').
3. The variable on disk is of type NC_BYTE, NC_CHAR, NC_SHORT, or NC_INT. Type NC_DOUBLE is not type converted because there is no type of higher precision to convert to. Type NC_FLOAT is not type converted because, in our judgement, the performance penalty of always doing so would outweigh the (extremely rare) potential benefits.

When these criteria are all met, the operator converts the variable in question to type NC_DOUBLE, performs all the arithmetic operations, casts the NC_DOUBLE type back to the original type, and finally writes the result to disk. The result written to disk may not be what you expect, because of incommensurate ranges represented by different types, and because of (lack of) rounding. First, continuing the example given above, the average (e.g., `-y avg') of 17000s and 17000s is written to disk as 17000s. The type conversion feature of NCO makes this possible since the arithmetic and intermediate values are stored as NC_DOUBLEs, i.e., 34000.0d and only the final result must be represented as an NC_SHORT. Without the type conversion feature of NCO, the average would have been garbage (albeit predictable garbage near -15768s). Similarly, the total (e.g., `-y ttl') of 17000s and 17000s written to disk is garbage (actually -31536s) since the final result (the true total) of 34000 is outside the range of type NC_SHORT.

Type conversions use the floor function to convert floating point number to integers. Type conversions do not attempt to round floating point numbers to the nearest integer. Thus the average of 1s and 2s is computed in double precisions arithmetic as (1.0d + 1.5d)/2) = 1.5d. This result is converted to NC_SHORT and stored on disk as floor(1.5d) = 1s (18). Thus no "rounding up" is performed. The type conversion rules of C can be stated as follows: If n is an integer then any floating point value x satisfying n <= x < n+1 will have the value n when converted to an integer.

3.13 Suppressing interactive prompts

3.14 History attribute

3.15 NCAR CSM Conventions

Currently, NCO determines whether a datafile is a CSM output datafile simply by checking whether value of the global attribute convention (if it exists) equals `NCAR-CSM'. Should convention equal `NCAR-CSM' in the (first) input-file, NCO will attempt to treat certain variables specially, because of their meaning in CSM files. NCO will not average the following variables often found in CSM files: ntrm, ntrn, ntrk, ndbase, nsbase, nbdate, nbsec, mdt, mhisf. These variables contain scalar metadata such as the resolution of the host CSM model and it makes no sense to change their values. Furthermore, the ncdiff operator will not attempt to difference the following variables: gw, ORO, date, datesec, hyam, hybm, hyai, hybi. These variables represent the Gaussian weights, the orography field, time fields, and hybrid pressure coefficients. These are fields which you want to remain unaltered in the differenced file 99% of the time. If you decide you would like any of the above CSM fields processed, you must use ncrename to rename them first.

3.16 ARM Conventions

3.17 Operator version

4. Reference manual for all operators

This chapter presents reference pages for each of the operators individually. The operators are presented in alphabetical order. All valid command line switches are included in the syntax statement. Recall that descriptions of many of these command line switches are provided only in 3. Features common to most operators, to avoid redundancy. Only options specific to, or most useful with, a particular operator are described in any detail in the sections below.

4.1 ncatted netCDF Attribute Editor
4.2 ncdiff netCDF Differencer
4.3 ncea netCDF Ensemble Averager
4.4 ncecat netCDF Ensemble Concatenator
4.5 ncflint netCDF File Interpolator
4.6 ncks netCDF Kitchen Sink
4.7 ncra netCDF Record Averager
4.8 ncrcat netCDF Record Concatenator
4.9 ncrename netCDF Renamer
4.10 ncwa netCDF Weighted Averager

4.1 ncatted netCDF Attribute Editor

SYNTAX

ncatted [-a att_dsc] [-a ...] [-D] [-h] [-l path] [-O] [-p path] [-R] [-r] input-file [output-file]

DESCRIPTION

ncatted edits attributes in a netCDF file. If you are editing attributes then you are spending too much time in the world of metadata, and ncatted was written to get you back out as quickly and painlessly as possible. ncatted can append, create, delete, modify, and overwrite attributes (all explained below). Furthermore, ncatted allows each editing operation to be applied to every variable in a file, thus saving you time when you want to change attribute conventions throughout a file. ncatted interprets character attributes as strings.

Because repeated use of ncatted can considerably increase the size of the history global attribute (see section 3.14 History attribute), the `-h' switch is provided to override automatically appending the command to the history global attribute in the output-file.

When ncatted is used to change the missing_value attribute, it changes the associated missing data self-consistently. If the internal floating point representation of a missing value, e.g., 1.0e36, differs between two machines then netCDF files produced on those machines will have incompatible missing values. This allows ncatted to change the missing values in files from different machines to a single value so that the files may then be concatenated together, e.g., by ncrcat, without losing any information. See section 3.10 Missing values, for more information.

The key to mastering ncatted is understanding the meaning of the structure describing the attribute modification, att_dsc. Each att_dsc contains five elements, which makes using ncatted somewhat complicated, but powerful. The att_dsc argument structure contains five arguments in the following order:

att_dsc = att_nm, var_nm, mode, att_type, att_val

att_nm

Attribute name. Example: units

var_nm

Variable name. Example: pressure

mode

Edit mode abbreviation. Example: a. See below for complete listing of valid values of mode.

att_type

Attribute type abbreviation. Example: c. See below for complete listing of valid values of att_type.

att_val

Attribute value. Example: pascal.

The value of att_nm is the name of the attribute you want to edit. This meaning of this should be clear to all users of the ncatted operator.

The value of var_nm is the name of the variable containing the attribute (named att_nm) that you want to edit. There are two very important and useful exceptions to this rule. The value of var_nm can also be used to direct ncatted to edit global attributes, or to repeat the editing operation for every variable in a file. A value of var_nm of "global" indicates that att_nm refers to a global attribute, rather than a particular variable's attribute. This is the method ncatted supports for editing global attributes. If var_nm is left blank, on the other hand, then ncatted attempts to perform the editing operation on every variable in the file. This option may be convenient to use if you decide to change the conventions you use for describing the data.

The value of mode is a single character abbreviation (a, c, d, m, or o) standing for one of five editing modes:

a

Append. Append value att_val to current var_nm attribute att_nm value att_val, if any. If var_nm does not have an attribute att_nm, there is no effect.

c

Create. Create variable var_nm attribute att_nm with att_val if att_nm does not yet exist. If var_nm already has an attribute att_nm, there is no effect.

d

Delete. Delete current var_nm attribute att_nm. If var_nm does not have an attribute att_nm, there is no effect. When Delete mode is selected, the att_type and att_val arguments are superfluous and may be left blank.

m

Modify. Change value of current var_nm attribute att_nm to value att_val. If var_nm does not have an attribute att_nm, there is no effect.

o

Overwrite. Write attribute att_nm with value att_val to variable var_nm, overwriting existing attribute att_nm, if any. This is the default mode.

The value of att_type is a single character abbreviation (f, d, l, i, s, c, or b) standing for one of the six primitive netCDF data types:

f

Float. Value(s) specified in att_val will be stored as netCDF intrinsic type NC_FLOAT.

d

Double. Value(s) specified in att_val will be stored as netCDF intrinsic type NC_DOUBLE.

i

Integer. Value(s) specified in att_val will be stored as netCDF intrinsic type NC_INT.

l

Long. Value(s) specified in att_val will be stored as netCDF intrinsic type NC_LONG.

s

Short. Value(s) specified in att_val will be stored as netCDF intrinsic type NC_SHORT.

c

Char. Value(s) specified in att_val will be stored as netCDF intrinsic type NC_CHAR.

b

Byte. Value(s) specified in att_val will be stored as netCDF intrinsic type NC_BYTE.

The value of att_val is what you want to change attribute att_nm to contain. The specification of att_val is optional in Delete mode. Attribute values for all types besides NC_CHAR must have an attribute length of at least one. Thus att_val may be a single value or one-dimensional array of elements of type att_type. If the att_val is not set or is set to empty space, and the att_type is NC_CHAR, e.g., -a units,T,o,c,"" or -a units,T,o,c,, then the corresponding attribute is set to have zero length. When specifying an array of values, it is safest to enclose att_val in double or single quotes, e.g., -a levels,T,o,s,"1,2,3,4" or -a levels,T,o,s,'1,2,3,4'. The quotes are strictly unnecessary around att_val except when att_val contains characters which would confuse the calling shell, such as spaces, commas, and wildcard characters.

NCO processing of NC_CHAR attributes is a bit like Perl in that it attempts to do what you want by default (but this sometimes causes unexpected results if you want unusual data storage). If the att_type is NC_CHAR then the argument is interpreted as a string and it may contain C-language escape sequences, e.g., \n, which NCO will interpret before writing anything to disk. NCO translates valid escape sequences and stores the appropriate ASCII code instead. Since two byte escape sequences, e.g., \n, represent one byte ASCII codes, e.g., ASCII 10 (decimal), the stored string attribute is one byte shorter than the input string length for each embedded escape sequence. The most frequently used C-language escape sequences are \n (for linefeed) and \t (for horizontal tab). These sequences in particular allow convenient editing of formatted text attributes. The other valid ASCII codes are \a, \b, \f, \r, \v, and \\. See section 4.6 ncks netCDF Kitchen Sink, for more examples of string formatting (with the ncks `-s' option) with special characters.

Analogous to printf, other special characters are also allowed by ncatted if they are "protected" by a backslash. The characters ", ', ?, and \ may be input to the shell as \", \', \?, and \\. NCO simply strips away the leading backslash from these characters before editing the attribute. No other characters require protection by a backslash. Backslashes which precede any other character (e.g., 3, m, $, |, &, @, %, {, and }) will not be filtered and will be included in the attribute.

Note that the NUL character \0 which terminates C language strings is assumed and need not be explicitly specified. If \0 is input, it will not be translated (because it would terminate the string in an additional location). Because of these context-sensitive rules, if wish to use an attribute of type NC_CHAR to store data, rather than text strings, you should use ncatted with care.

EXAMPLES

Append the string "Data version 2.0.\n" to the global attribute history:

ncatted -O -a history,global,a,c,"Data version 2.0\n" in.nc

Note the use of embedded C language printf()-style escape sequences.

Change the value of the long_name attribute for variable T from whatever it currently is to "temperature":

ncatted -O -a long_name,T,o,c,temperature in.nc

Delete all existing units attributes:

ncatted -O -a units,,d,, in.nc

The value of var_nm was left blank in order to select all variables in the file. The values of att_type and att_val were left blank because they are superfluous in Delete mode.

Modify all existing units attributes to "meter second-1"

ncatted -O -a units,,m,c,"meter second-1" in.nc

Overwrite the quanta attribute of variable energy to an array of four integers.

ncatted -O -a quanta,energy,o,s,"010,101,111,121" in.nc

Demonstrate input of C-language escape sequences (e.g., \n) and other special characters (e.g., \")

ncatted -h -a special,global,o,c, '\nDouble quote: \"\nTwo consecutive double quotes: \"\"\n Single quote: Beyond my shell abilities!\nBackslash: \\\n Two consecutive backslashes: \\\\\nQuestion mark: \?\n' in.nc

Note that the entire attribute is protected from the shell by single quotes. These outer single quotes are necessary for interactive use, but may be omitted in batch scripts.

4.2 ncdiff netCDF Differencer

SYNTAX

ncdiff [-A] [-C] [-c] [-D dbg] [-d dim,[min][,[max]]] [-F] [-h] [-l path] [-O] [-p path] [-R] [-r] [-v var[,...]] [-x] file_1 file_2 file_3

DESCRIPTION

ncdiff subtracts variables in file_2 from the corresponding variables (those with the same name) in file_1 and stores the results in file_3. Variables in file_2 are broadcast to conform to the corresponding variable in file_1 if necessary. Broadcasting a variable means creating data in non-existing dimensions from the data in existing dimensions. For example, a two dimensional variable in file_2 can be subtracted from a four, three, or two (but not one or zero) dimensional variable (of the same name) in file_1. This functionality allows the user to compute anomalies from the mean. Note that variables in file_1 are not broadcast to conform to the dimensions in file_2. Thus, ncdiff, the number of dimensions, or rank, of any processed variable in file_1 must be greater than or equal to the rank of the same variable in file_2. Furthermore, the size of all dimensions common to both file_1 and file_2 must be equal.

When computing anomalies from the mean it is often the case that file_2 was created by applying an averaging operator to a file with the same dimensions as file_1, if not file_1 itself. In these cases, creating file_2 with ncra rather than ncwa will cause the ncdiff operation to fail. For concreteness say the record dimension in file_1 is time. If file_2 were created by averaging file_1 over the time dimension with the ncra operator rather than with the ncwa operator, then file_2 will have a time dimension of size 1 rather than having no time dimension at all (21). In this case the input files to ncdiff, file_1 and file_2, will have unequally sized time dimensions which causes ncdiff to fail. To prevent this from occuring, use ncwa to remove the time dimension from file_2. An example is given below.

ncdiff will never difference coordinate variables or variables of type NC_CHAR or NC_BYTE. This ensures that coordinates like (e.g., latitude and longitude) are physically meaningful in the output file, file_3. This behavior is hardcoded. ncdiff applies special rules to some NCAR CSM fields (e.g., ORO). See 3.15 NCAR CSM Conventions for a complete description. Finally, we note that ncflint (see section 4.5 ncflint netCDF File Interpolator) can be also perform file subtraction (as well as addition, multiplication and interpolation).

EXAMPLES

Say files `85_0112.nc' and `86_0112.nc' each contain 12 months of data. Compute the change in the monthly averages from 1985 to 1986:

ncdiff 86_0112.nc 85_0112.nc 86m85_0112.nc

The following examples demonstrate the broadcasting feature of ncdiff. Say we wish to compute the monthly anomalies of T from the yearly average of T for the year 1985. First we create the 1985 average from the monthly data, which is stored with the record dimension time.

ncra 85_0112.nc 85.nc ncwa -O -a time 85.nc 85.nc

The second command, ncwa, gets rid of the time dimension of size 1 that ncra left in `85.nc'. Now none of the variables in `85.nc' has a time dimension. A quicker way to accomplish this is to use ncwa from the beginning:

ncwa -a time 85_0112.nc 85.nc

We are now ready to use ncdiff to compute the anomalies for 1985:

ncdiff -v T 85_0112.nc 85.nc t_anm_85_0112.nc

Each of the 12 records in `t_anm_85_0112.nc' now contains the monthly deviation of T from the annual mean of T for each gridpoint.

Say we wish to compute the monthly gridpoint anomalies from the zonal annual mean. A zonal mean is a quantity that has been averaged over the longitudinal (or x) direction. First we use ncwa to average over longitudinal direction lon, creating `xavg_85.nc', the zonal mean of `85.nc'. Then we use ncdiff to subtract the zonal annual means from the monthly gridpoint data:

ncwa -a lon 85.nc xavg_85.nc ncdiff 85_0112.nc xavg_85.nc tx_anm_85_0112.nc

Assuming `85_0112.nc' has dimensions time and lon, this example only works if `xavg_85.nc' has no time or lon dimension.

As a final example, say we have five years of monthly data (i.e., 60 months) stored in `8501_8912.nc' and we wish to create a file which contains the twelve month seasonal cycle of the average monthly anomaly from the five-year mean of this data. The following method is just one permutation of many which will accomplish the same result. First use ncwa to create the file containing the five-year mean:

ncwa -a time 8501_8912.nc 8589.nc

Next use ncdiff to create a file containing the difference of each month's data from the five-year mean:

ncdiff 8501_8912.nc 8589.nc t_anm_8501_8912.nc

Now use ncks to group the five January anomalies together in one file, and use ncra to create the average anomaly for all five Januarys. These commands are embedded in a shell loop so they are repeated for all twelve months:

foreach idx (01 02 03 04 05 06 07 08 09 10 11 12) ncks -F -d time,$idx,,12 t_anm_8501_8912.nc foo.$idx ncra foo.$idx t_anm_8589_$idx.nc end

Note that ncra understands the stride argument so the two commands inside the loop may be combined into the single command

ncra -F -d time,$idx,,12 t_anm_8501_8912.nc foo.$idx

Finally, use ncrcat to concatenate the 12 average monthly anomaly files into one twelve-record file which contains the entire seasonal cycle of the monthly anomalies:

ncrcat t_anm_8589_??.nc t_anm_8589_0112.nc

4.3 ncea netCDF Ensemble Averager

SYNTAX

ncea [-A] [-C] [-c] [-D dbg] [-d dim,[min][,[max]]] [-F] [-h] [-l path] [-n loop] [-O] [-p path] [-R] [-r] [-v var[,...]] [-x] [-y op_typ] input-files output-file

DESCRIPTION

ncea performs gridpoint averages of variables across an arbitrary number (an ensemble) of input files, with each file receiving an equal weight in the average. Each variable in the output-file will be the same size as the same variable in any one of the in the input-files, and all input-files must be the same size. Whereas ncra only performs averages over the record dimension (e.g., time), and weights each record in the record dimension evenly, ncea averages entire files, and weights each file evenly. All dimensions, including the record dimension, are treated identically and preserved in the output-file. See section 2.6 Averagers vs. Concatenators, for a description of the distinctions between the various averagers and concatenators.

The file is the logical unit of organization for the results of many scientific studies. Often one wishes to generate a file which is the gridpoint average of many separate files. This may be to reduce statistical noise by combining the results of a large number of experiments, or it may simply be a step in a procedure whose goal is to compute anomalies from a mean state. In any case, when one desires to generate a file whose properties are the mean of all the input files, then ncea is the operator to use. ncea assumes coordinate variable are properties common to all of the experiments and so does not average them across files. Instead, ncea copies the values of the coordinate variables from the first input file to the output file.

EXAMPLES

Consider a model experiment which generated five realizations of one year of data, say 1985. You can imagine that the experimenter slightly perturbs the initial conditions of the problem before generating each new solution. Assume each file contains all twelve months (a seasonal cycle) of data and we want to produce a single file containing the ensemble average (mean) seasonal cycle. Here the numeric filename suffix denotes the experiment number (not the month):

ncea 85_01.nc 85_02.nc 85_03.nc 85_04.nc 85_05.nc 85.nc ncea 85_0[1-5].nc 85.nc ncea -n 5,2,1 85_01.nc 85.nc

These three commands produce identical answers. See section 3.1 Specifying input files, for an explanation of the distinctions between these methods. The output file, `85.nc', is the same size as the inputs files. It contains 12 months of data (which might or might not be stored in the record dimension, depending on the input files), but each value in the output file is the average of the five values in the input files.

In the previous example, the user could have obtained the ensemble average values in a particular spatio-temporal region by adding a hyperslab argument to the command, e.g.,

ncea -d time,0,2 -d lat,-23.5,23.5 85_??.nc 85.nc

In this case the output file would contain only three slices of data in the time dimension. These three slices are the average of the first three slices from the input files. Additionally, only data inside the tropics is included.

4.4 ncecat netCDF Ensemble Concatenator

SYNTAX

ncecat [-A] [-C] [-c] [-D dbg] [-d dim,[min][,[max]]] [-F] [-h] [-l path] [-n loop] [-O] [-p path] [-R] [-r] [-v var[,...]] [-x] input-files output-file

DESCRIPTION

ncecat concatenates an arbitrary number of input files into a single output file. Input files are glued together by creating a record dimension in the output file. Input files must be the same size. Each input file is stored consecutively as a single record in the output file. Thus, the size of the output file is the sum of the sizes of the input files. See section 2.6 Averagers vs. Concatenators, for a description of the distinctions between the various averagers and concatenators.

Consider five realizations, `85a.nc', `85b.nc', ... `85e.nc' of 1985 predictions from the same climate model. Then ncecat 85?.nc 85_ens.nc glues the individual realizations together into the single file, `85_ens.nc'. If an input variable was dimensioned [lat,lon], it will have dimensions [record,lat,lon] in the output file. A restriction of ncecat is that the hyperslabs of the processed variables must be the same from file to file. Normally this means all the input files are the same size, and contain data on different realizations of the same variables.

EXAMPLES

Consider a model experiment which generated five realizations of one year of data, say 1985. You can imagine that the experimenter slightly perturbs the initial conditions of the problem before generating each new solution. Assume each file contains all twelve months (a seasonal cycle) of data and we want to produce a single file containing all the seasonal cycles. Here the numeric filename suffix denotes the experiment number (not the month):

ncecat 85_01.nc 85_02.nc 85_03.nc 85_04.nc 85_05.nc 85.nc ncecat 85_0[1-5].nc 85.nc ncecat -n 5,2,1 85_01.nc 85.nc

These three commands produce identical answers. See section 3.1 Specifying input files, for an explanation of the distinctions between these methods. The output file, `85.nc', is five times the size as a single input-file. It contains 60 months of data (which might or might not be stored in the record dimension, depending on the input files).

4.5 ncflint netCDF File Interpolator

SYNTAX

ncflint [-A] [-C] [-c] [-D dbg] [-d dim,[min][,[max]]] [-F] [-h] [-i var,val3] [-l path] [-O] [-p path] [-R] [-r] [-v var[,...]] [-w wgt1[,wgt2]] [-x] file_1 file_2 file_3

DESCRIPTION

ncflint creates an output file that is a linear combination of the input files. This linear combination can be a weighted average, a normalized weighted average, or an interpolation of the input files. Coordinate variables are not acted upon in any case, they are simply copied from file_1.

There are two conceptually distinct methods of using ncflint. The first method is to specify the weight each input file is to have in the output file. In this method, the value val3 of a variable in the output file file_3 is determined from its values val1 and val2 in the two input files according to val3 = wgt1*val1 + wgt2*val2 . Here at least wgt1, and, optionally, wgt2, are specified on the command line with the `-w' switch. If only wgt1 is specified then wgt2 is automatically computed as wgt2 = 1 - wgt1. Note that weights larger than 1 are allowed. Thus it is possible to specify wgt1 = 2 and wgt2 = -3. One can use this functionality to multiply all the values in a given file by a constant.

The second method of using ncflint is specifying the interpolation option with `-i'. This is really the inverse of the first method in the following sense. When the user specifies the weights directly, ncflint has no work to do besides multiplying the input values by their respective weights and adding the results together to produce the output values. This assumes it is the weights that are known a priori. In another class of cases it is the arrival value (i.e., val3) of a particular variable var that is known a priori. In this case, the implied weights can always be inferred by examining the values of var in the input files. This results in one equation in two unknowns, wgt1 and wgt2: val3 = wgt1*val1 + wgt2*val2 . Unique determination of the weights requires imposing the additional constraint of normalization on the weights: wgt1 + wgt2 = 1. Thus, to use the interpolation option, the user specifies var and val3 with the `-i' option. ncflint will compute wgt1 and wgt2, and use these weights on all variables to generate the output file. Although var may have any number of dimensions in the input files, it must represent a single, scalar value. Thus any dimensions associated with var must be degenerate, i.e., of size one.

If neither `-i' nor `-w' is specified on the command line, ncflint defaults to weighting each input file equally in the output file. This is equivalent to specifying `-w 0.5' or `-w 0.5,0.5'. Attempting to specify both `-i' and `-w' methods in the same command is an error.

ncflint is programmed not to interpolate variables of type NC_CHAR and NC_BYTE. This behavior is hardcoded.

EXAMPLES

Although it has other uses, the interpolation feature was designed to interpolate file_3 to a time between existing files. Consider input files `85.nc' and `87.nc' containing variables describing the state of a physical system at times time = 85 and time = 87. Assume each file contains its timestamp in the scalar variable time. Then, to linearly interpolate to a file `86.nc' which describes the state of the system at time at time = 86, we would use

ncflint -i time,86 85.nc 87.nc 86.nc

Say you have observational data covering January and April 1985 in two files named `85_01.nc' and `85_04.nc', respectively. Then you can estimate the values for February and March by interpolating the existing data as follows. Combine `85_01.nc' and `85_04.nc' in a 2:1 ratio to make `85_02.nc':

ncflint -w 0.667 85_01.nc 85_04.nc 85_02.nc ncflint -w 0.667,0.333 85_01.nc 85_04.nc 85_02.nc

Multiply `85.nc' by 3 and by -2 and add them together to make `tst.nc':

ncflint -w 3,-2 85.nc 85.nc tst.nc

This is an example of a null operation, so `tst.nc' should be identical (within machine precision) to `85.nc'.

Add `85.nc' to `86.nc' to obtain `85p86.nc', then subtract `86.nc' from `85.nc' to obtain `85m86.nc'

ncflint -w 1,1 85.nc 86.nc 85p86.nc ncflint -w 1,-1 85.nc 86.nc 85m86.nc ncdiff 85.nc 86.nc 85m86.nc

Thus ncflint can be used to mimic ncdiff operations. However this is not a good idea in practice because ncflint does not broadcast (see section 4.2 ncdiff netCDF Differencer) conforming variables during arithmetic. Thus the final two commands would produce identical results except that ncflint would fail if any variables needed to be broadcast.

Rescale the dimensional units of the surface pressure prs_sfc from Pascals to hectopascals (millibars)

ncflint -O -C -v prs_sfc -w 0.01,0.0 in.nc in.nc out.nc ncatted -O -a units,prs_sfc,o,c,millibar out.nc

4.6 ncks netCDF Kitchen Sink

SYNTAX

ncks [-A] [-a] [-C] [-c] [-D] [-d dim,[min][,[max]][,[stride]]] [-F] [-H] [-h] [-l path] [-M] [-m] [-O] [-p path] [-q] [-R] [-r] [-s format] [-u] [-v var[,...]] [-x] input-file [output-file]

DESCRIPTION

ncks combines selected features of ncdump, ncextr, and the nccut and ncpaste specifications into one versatile utility. ncks extracts a subset of the data from input-file and either prints it as ASCII text to stdout, or writes (or pastes) it to output-file, or both.

ncks will print netCDF data in ASCII format to stdout, like ncdump, but with these differences: ncks prints data in a tabular format intended to be easy to search for the data you want, one datum per screen line, with all dimension subscripts and coordinate values (if any) preceding the datum. Option `-s' allows the user the format the data using C-style format strings.

Options `-a', `-F', `-H', `-M', `-m', `-q', `-s', and `-u' control the formatted appearance of the data.

ncks will extract (and optionally create a new netCDF file comprised of) only selected variable from the input file, like ncextr but with these differences: Only variables and coordinates may be specifically included or excluded--all global attributes and any attribute associated with an extracted variable will be copied to the screen and/or output netCDF file. Options `-c', `-C', `-v', and `-x' control which variables are extracted.

ncks will extract hyperslabs from the specified variables. In fact ncks implements the nccut specification exactly. Option `-d' controls the hyperslab specification.

Input dimensions that are not associated with any output variable will not appear in the output netCDF. This feature removes superfluous dimensions from a netCDF file.

ncks will append variables and attributes from the input-file to output-file if output-file is a pre-existing netCDF file whose relevant dimensions conform to dimension sizes of input-file. The append features of ncks are intended to provide a rudimentary means of adding data from one netCDF file to another, conforming, netCDF file. When naming conflicts exists between the two files, data in output-file is usually overwritten by the corresponding data from input-file. Thus it is recommended that the user backup output-file in case valuable data is accidentally overwritten.

If output-file exists, the user will be queried whether to overwrite, append, or exit the ncks call completely. Choosing overwrite destroys the existing output-file and create an entirely new one from the output of the ncks call. Append has differing effects depending on the uniqueness of the variables and attributes output by ncks: If a variable or attribute extracted from input-file does not have a name conflict with the members of output-file then it will be added to output-file without overwriting any of the existing contents of output-file. In this case the relevant dimensions must agree (conform) between the two files; new dimensions are created in output-file as required. When a name conflict occurs, a global attribute from input-file will overwrite the corresponding global attribute from output-file. If the name conflict occurs for a non-record variable, then the dimensions and type of the variable (and of its coordinate dimensions, if any) must agree (conform) in both files. Then the variable values (and any coordinate dimension values) from input-file will overwrite the corresponding variable values (and coordinate dimension values, if any) in output-file (22).

Since there can only be one record dimension in a file, the record dimension must have the same name (but not necessarily the same size) in both files if a record dimension variable is to be appended. If the record dimensions are of differing sizes, the record dimension of output-file will become the greater of the two record dimension sizes, the record variable from input-file will overwrite any counterpart in output-file and fill values will be written to any gaps left in the rest of the record variables (I think). In all cases variable attributes in output-file are superseded by attributes of the same name from input-file, and left alone if there is no name conflict.

Some users may wish to avoid interactive ncks queries about whether to overwrite existing data. For example, batch scripts will fail if ncks does not receive responses to its queries. Options `-O' and `-A' are available to force overwriting existing files and variables, respectively.

Options specific to ncks

The following list provides a short summary of the features unique to ncks. Features common to many operators are described in 3. Features common to most operators.

`-a'

Do not alphabetize extracted fields. By default, the specified output variables are extracted, printed, and written to disk in alphabetical order. This tends to make long output lists easier to search for particular variables. Specifying -a results in the variables being extracted, printed, and written to disk in the order in which they were saved in the input file. Thus -a retains the original ordering of the variables.

`-d dim,[min][,[max]][,[stride]]'

Add stride argument to hyperslabber. For a complete description of the stride argument, See section 3.9 Stride.

`-H'

Print data to screen. Unless otherwise specified (with -s), each element of the data hyperslab is printed on a separate line containing the names, indices, and, values, if any, of all of the variables dimensions. The dimension and variable indices refer to the location of the corresponding data element with respect to the variable as stored on disk (i.e., not the hyperslab).

% ncks -H -C -v three_dmn_var in.nc lat[0]=-90 lev[0]=100 lon[0]=0 three_dmn_var[0]=0 lat[0]=-90 lev[0]=100 lon[1]=90 three_dmn_var[1]=1 lat[0]=-90 lev[0]=100 lon[2]=180 three_dmn_var[2]=2 ... lat[1]=90 lev[2]=1000 lon[1]=90 three_dmn_var[21]=21 lat[1]=90 lev[2]=1000 lon[2]=180 three_dmn_var[22]=22 lat[1]=90 lev[2]=1000 lon[3]=270 three_dmn_var[23]=23

Printing the same variable with the `-F' option shows the same variable indexed with Fortran conventions

% ncks -F -H -C -v three_dmn_var in.nc lon(1)=0 lev(1)=100 lat(1)=-90 three_dmn_var(1)=0 lon(2)=90 lev(1)=100 lat(1)=-90 three_dmn_var(2)=1 lon(3)=180 lev(1)=100 lat(1)=-90 three_dmn_var(3)=2 ...

Printing a hyperslab does not affect the variable or dimension indices since these indices are relative to the full variable (as stored in the input file), and the input file has not changed. However, if the hypserslab is saved to an output file and those values are printed, the indices will change:

% ncks -H -d lat,90.0 -d lev,1000.0 -v three_dmn_var in.nc out.nc lat[1]=90 lev[2]=1000 lon[0]=0 three_dmn_var[20]=20 lat[1]=90 lev[2]=1000 lon[1]=90 three_dmn_var[21]=21 lat[1]=90 lev[2]=1000 lon[2]=180 three_dmn_var[22]=22 lat[1]=90 lev[2]=1000 lon[3]=270 three_dmn_var[23]=23 % ncks -H out.nc lat[0]=90 lev[0]=1000 lon[0]=0 three_dmn_var[0]=20 lat[0]=90 lev[0]=1000 lon[1]=90 three_dmn_var[1]=21 lat[0]=90 lev[0]=1000 lon[2]=180 three_dmn_var[2]=22 lat[0]=90 lev[0]=1000 lon[3]=270 three_dmn_var[3]=23

`-M'

Print to screen the global metadata describing the file. This includes file summary information and global attributes.

`-m'

Print variable metadata to screen (similar to ncdump -h). This displays all metadata pertaining to each variable, one variable at a time.

`-q'

Toggle printing of dimension indices and coordinate values when printing arrays. The name of each variable will appear flush left in the output. This is useful when trying to locate specific variables when displaying many variables with different dimensions. The mnemonic for this option is "quiet".

`-s format'

String format for text output. Accepts C language escape sequences and printf() formats.

`-u'

Accompany the printing of a variable's values with its units attribute, if it exists.

EXAMPLES

View all data in netCDF `in.nc', printed with Fortran indexing conventions:

ncks -H -F in.nc

Copy the netCDF file `in.nc' to file `out.nc'.

ncks -O in.nc out.nc

Now the file `out.nc' contains all the data from `in.nc'. There are, however, two differences between `in.nc' and `out.nc'. First, the history global attribute (see section 3.14 History attribute) will contain the command used to create `out.nc'. Second, the variables in `out.nc' will be defined in alphabetical order. Of course the internal storage of variable in a netCDF file should be transparent to the user, but there are cases when alphabetizing a file is useful (see description of -a switch).

Print variable three_dmn_var from file `in.nc' with default notations. Next print three_dmn_var as an un-annotated text column. Then print three_dmn_var signed with very high precision. Finally, print three_dmn_var as a comma-separated list.

% ncks -H -C -v three_dmn_var in.nc lat[0]=-90 lev[0]=100 lon[0]=0 three_dmn_var[0]=0 lat[0]=-90 lev[0]=100 lon[1]=90 three_dmn_var[1]=1 ... lat[1]=90 lev[2]=1000 lon[3]=270 three_dmn_var[23]=23 % ncks -s "%f\n" -H -C -v three_dmn_var in.nc 0.000000 1.000000 ... 23.000000 % ncks -s "%+16.10f\n" -H -C -v three_dmn_var in.nc +0.0000000000 +1.0000000000 ... +23.0000000000 % ncks -s "%f, " -H -C -v three_dmn_var in.nc 0.000000, 1.000000, ..., 23.000000,

The second and third options are useful when pasting data into text files like reports or papers. See section 4.1 ncatted netCDF Attribute Editor, for more details on string formatting and special characters.

One dimensional arrays of characters stored as netCDF variables are automatically printed as strings, whether or not they are NUL-terminated, e.g.,

ncks -v fl_nm in.nc

The %c formatting code is useful for printing multidimensional arrays of characters representing fixed length strings

ncks -H -s "%c" -v fl_nm_arr in.nc

Using the %s format code on strings which are not NUL-terminated (and thus not technically strings) is likely to result in a core dump.

Create netCDF `out.nc' containing all variables, and any associated coordinates, except variable time, from netCDF `in.nc':

ncks -x -v time in.nc out.nc

Extract variables time and pressure from netCDF `in.nc'. If `out.nc' does not exist it will be created. Otherwise the you will be prompted whether to append to or to overwrite `out.nc':

ncks -v time,pressure in.nc out.nc ncks -C -v time,pressure in.nc out.nc

The first version of the command creates an `out.nc' which contains time, pressure, and any coordinate variables associated with pressure. The `out.nc' from the second version is guaranteed to contain only two variables time and pressure.

Create netCDF `out.nc' containing all variables from file `in.nc'. Restrict the dimensions of these variables to a hyperslab. Print (with -H) the hyperslabs to the screen for good measure. The specified hyperslab is: the fifth value in dimension time; the half-open range lat > 0. in coordinate lat; the half-open range lon < 330. in coordinate lon; the closed interval .3 < band < .5 in coordinate band; and cross-section closest to 1000. in coordinate lev. Note that limits applied to coordinate values are specified with a decimal point, and limits applied to dimension indices do not have a decimal point See section 3.7 Hyperslabs.

ncks -H -d time,5 -d lat,,0. -d lon,330., -d band,.3,.5 -d lev,1000. in.nc out.nc

Assume the domain of the monotonically increasing longitude coordinate lon is 0 < lon < 360. Here, lon is an example of a wrapped coordinate. ncks will extract a hyperslab which crosses the Greenwich meridian simply by specifying the westernmost longitude as min and the easternmost longitude as max, as follows:

ncks -d lon,260.,45. in.nc out.nc

For more details See section 3.8 Wrapped coordinates.

4.7 ncra netCDF Record Averager

SYNTAX

ncra [-A] [-C] [-c] [-D dbg] [-d dim,[min][,[max]][,[stride]]] [-F] [-h] [-l path] [-n loop] [-O] [-p path] [-R] [-r] [-v var[,...]] [-x] [-y op_typ] input-files output-file

DESCRIPTION

ncra averages record variables across an arbitrary number of input files. The record dimension is retained as a degenerate (size 1) dimension in the output variables. See section 2.6 Averagers vs. Concatenators, for a description of the distinctions between the various averagers and concatenators.

Input files may vary in size, but each must have a record dimension. The record coordinate, if any, should be monotonic for (or else non-fatal warnings may be generated). Hyperslabs of the record dimension which include more than one file are handled correctly. ncra supports the stride argument to the `-d' hyperslab option for the record dimension only, stride is not supported for non-record dimensions.

ncra weights each record (e.g., time slice) in the input-files equally. ncra does not attempt to see if, say, the time coordinate is irregularly spaced and thus would require a weighted average in order to be a true time average.

EXAMPLES

Average files `85.nc', `86.nc', ... `89.nc' along the record dimension, and store the results in `8589.nc':

ncra 85.nc 86.nc 87.nc 88.nc 89.nc 8589.nc ncra 8[56789].nc 8589.nc ncra -n 5,2,1 85.nc 8589.nc

These three methods produce identical answers. See section 3.1 Specifying input files, for an explanation of the distinctions between these methods.

Assume the files `85.nc', `86.nc', ... `89.nc' each contain a record coordinate time of length 12 defined such that the third record in `86.nc' contains data from March 1986, etc. NCO knows how to hyperslab the record dimension across files. Thus, to average data from December, 1985 through February, 1986:

ncra -d time,11,13 85.nc 86.nc 87.nc 8512_8602.nc ncra -F -d time,12,14 85.nc 86.nc 87.nc 8512_8602.nc

The file `87.nc' is superfluous, but does not cause an error. The `-F' turns on the Fortran (1-based) indexing convention. The following uses the stride option to average all the March temperature data from multiple input files into a single output file

ncra -F -d time,3,,12 -v temperature 85.nc 86.nc 87.nc 858687_03.nc

See section 3.9 Stride, for a description of the stride argument.

Assume the time coordinate is incrementally numbered such that January, 1985 = 1 and December, 1989 = 60. Assuming `??' only expands to the five desired files, the following averages June, 1985--June, 1989:

ncra -d time,6.,54. ??.nc 8506_8906.nc

4.8 ncrcat netCDF Record Concatenator

SYNTAX

ncrcat [-A] [-C] [-c] [-D dbg] [-d dim,[min][,[max]][,[stride]]] [-F] [-h] [-l path] [-n loop] [-O] [-p path] [-R] [-r] [-v var[,...]] [-x] input-files output-file

DESCRIPTION

ncrcat concatenates record variables across an arbitrary number of input files. The final record dimension is by default the sum of the lengths of the record dimensions in the input files. See section 2.6 Averagers vs. Concatenators, for a description of the distinctions between the various averagers and concatenators.

Input files may vary in size, but each must have a record dimension. The record coordinate, if any, should be monotonic (or else non-fatal warnings may be generated). Hyperslabs of the record dimension which include more than one file are handled correctly. ncra supports the stride argument to the `-d' hyperslab option for the record dimension only, stride is not supported for non-record dimensions.

ncrcat applies special rules to ARM convention time fields (e.g., time_offset). See 3.16 ARM Conventions for a complete description.

EXAMPLES

Concatenate files `85.nc', `86.nc', ... `89.nc' along the record dimension, and store the results in `8589.nc':

ncrcat 85.nc 86.nc 87.nc 88.nc 89.nc 8589.nc ncrcat 8[56789].nc 8589.nc ncrcat -n 5,2,1 85.nc 8589.nc

These three methods produce identical answers. See section 3.1 Specifying input files, for an explanation of the distinctions between these methods.

Assume the files `85.nc', `86.nc', ... `89.nc' each contain a record coordinate time of length 12 defined such that the third record in `86.nc' contains data from March 1986, etc. NCO knows how to hyperslab the record dimension across files. Thus, to concatenate data from December, 1985--February, 1986:

ncrcat -d time,11,13 85.nc 86.nc 87.nc 8512_8602.nc ncrcat -F -d time,12,14 85.nc 86.nc 87.nc 8512_8602.nc

The file `87.nc' is superfluous, but does not cause an error. The `-F' turns on the Fortran (1-based) indexing convention.

The following uses the stride option to concatenate all the March temperature data from multiple input files into a single output file

ncrcat -F -d time,3,,12 -v temperature 85.nc 86.nc 87.nc 858687_03.nc

See section 3.9 Stride, for a description of the stride argument.

Assume the time coordinate is incrementally numbered such that January, 1985 = 1 and December, 1989 = 60. Assuming ?? only expands to the five desired files, the following concatenates June, 1985--June, 1989:

ncrcat -d time,6.,54. ??.nc 8506_8906.nc

4.9 ncrename netCDF Renamer

SYNTAX

ncrename [-a old_name,new_name] [-a ...] [-D] [-d old_name,new_name] [-d ...] [-h] [-l path] [-O] [-p path] [-R] [-r] [-v old_name,new_name] [-v ...] input-file [output-file]

DESCRIPTION

ncrename renames dimensions, variables, and attributes in a netCDF file. Each object that has a name in the list of old names is renamed using the corresponding name in the list of new names. All the new names must be unique. Every old name must exist in the input file, unless the name is preceded by the character `.'. The validity of the old names is not checked prior to the renaming. Thus, if an old name is specified without the the `.' prefix and is not present in input-file, ncrename will abort.

ncrename is the exception to the normal rules that the user will be interactively prompted before an existing file is changed, and that a temporary copy of an output file is constructed during the operation. If only input-file is specified, then ncrename will change the names of the input-file in place without prompting and without creating a temporary copy of input-file. This is because the renaming operation is considered reversible if the user makes a mistake. The new_name can easily be changed back to old_name by using ncrename one more time.

Note that renaming a dimension to the name of a dependent variable can be used to invert the relationship between an independent coordinate variable and a dependent variable. In this case, the named dependent variable must be one-dimensional and should have no missing values. Such a variable will become a coordinate variable.

According to the netCDF User's Guide, renaming properties in netCDF files does not incur the penalty of recopying the entire file when the new_name is shorter than the old_name.

OPTIONS

`-a old_name,new_name'

Attribute renaming. The old and new names of the attribute are specified by the associated old_name and new_name values. Global attributes are treated no differently than variable attributes. This option may be specified more than once. You cannot change the attribute name for one particular variable (unless it is uniquely named); all occurrences of the attribute of a given name will be renamed. This is considered an oversight and will be addressed in a future version of NCO.

`-d old_name,new_name'

Dimension renaming. The old and new names of the dimension are specified by the associated old_name and new_name values. This option may be specified more than once.

`-v old_name,new_name'

Variable renaming. The old and new names of the variable are specified by the associated old_name and new_name values. This option may be specified more than once.

EXAMPLES

Rename the variable p to pressure and t to temperature in netCDF `in.nc'. In this case p must exist in the input file (or ncrename will abort), but the presence of t is optional:

ncrename -v p,pressure -v .t,temperature in.nc

ncrename does not automatically attach dimensions to variables of the same name. If you want to rename a coordinate variable so that it remains a coordinate variable, you must separately rename both the dimension and the variable:

ncrename -d lon,longitude -v lon,longitude in.nc

Create netCDF `out.nc' identical to `in.nc' except the attribute _FillValue is changed to missing_value (in all variables which possess it) and the global attribute Zaire is changed to Congo:

ncrename -a _FillValue,missing_value -a Zaire,Congo in.nc out.nc

4.10 ncwa netCDF Weighted Averager

SYNTAX

ncwa [-A] [-a dim[,...]] [-C] [-c] [-D dbg] [-d dim,[min][,[max]]] [-F] [-h] [-I] [-l path] [-M val] [-m mask] [-N] [-n] [-O] [-o condition] [-p path] [-R] [-r] [-v var[,...]] [-W] [-w weight] [-x] [-y op_typ] input-file output-file

DESCRIPTION

ncwa averages variables in a single file over arbitrary dimensions, with options to specify weights, masks, and normalization. See section 2.6 Averagers vs. Concatenators, for a description of the distinctions between the various averagers and concatenators. The default behavior of ncwa is to arithmetically average every numerical variable over all dimensions and produce a scalar result. To average variables over only a subset of their dimensions, specify these dimensions in a comma-separated list following `-a', e.g., `-a time,lat,lon'. As with all arithmetic operators, the operation may be restricted to an arbitrary hypserslab by employing the `-d' option (see section 3.7 Hyperslabs). ncwa also handles values matching the variable's missing_value attribute correctly. Moreover, ncwa understands how to manipulate user-specified weights, masks, and normalization options. With these options, ncwa can compute sophisticated averages (and integrals) from the command line.

mask and weight, if specified, are broadcast to conform to the variables being averaged. The rank of variables is reduced by the number of dimensions which they are averaged over. Thus arrays which are one dimensional in the input-file and are averaged by ncwa appear in the output-file as scalars. This allows the user to infer which dimensions may have been averaged. Note that that it is impossible for ncwa to make make a weight or mask of rank W conform to a var of rank V if W > V. This situation often arises when coordinate variables (which, by definition, are one dimensional) are weighted and averaged. ncwa assumes you know this is impossible and so ncwa does not attempt to broadcast weight or mask to conform to var in this case, nor does ncwa print a warning message telling you this, because it is so common. Specifying dbg > 2 does cause ncwa to emit warnings in these situations, however.

Non-coordinate variables are always masked and weighted if specified. Coordinate variables, however, may be treated specially. By default, an averaged coordinate variable, e.g., latitude, appears in output-file averaged the same way as any other variable containing an averaged dimension. In other words, by default ncwa weights and masks coordinate variables like all other variables. This design decision was intended to be helpful but for some applications it may be preferable not to weight or mask coordinate variables just like all other variables. Consider the following arguments to ncwa: -a latitude -w lat_wgt -d latitude,0.,90. where lat_wgt is a weight in the latitude dimension. Since, by default ncwa weights coordinate variables, the value of latitude in the output-file depends on the weights in lat_wgt and is not likely to be 45.---the midpoint latitude of the hyperslab. Option `-I' overrides this default behavior and causes ncwa not to weight or mask coordinate variables. In the above case, this causes the value of latitude in the output-file to be 45.---which is a somewhat appealing result. Thus, `-I' specifies simple arithmetic averages for the coordinate variables. In the case of latitude, `-I' specifies that you prefer to archive the central latitude of the hyperslab over which variables were averaged rather than the area weighted centroid of the hyperslab (23). Note that the default behavior of (`-I') changed on 1998/12/01--before this date the default was not to weight or mask coordinate variables. The mathematical definition of operations involving rank reduction is given above (see section 3.11 Operation Types).

Masking condition
Normalization

Masking condition

The masking condition has the syntax mask condition val. Here mask is the name of the masking variable (specified with `-m'). The condition argument to `-o' may be any one of the six arithmetic comparatives: eq, ne, gt, lt, ge, le. These are the Fortran-style character abbreviations for the logical operations ==, !=, >, <, >=, <=. The masking condition defaults to eq (equality). The val argument to `-M' is the right hand side of the masking condition. Thus for the i'th element of the hyperslab to be averaged, the masking condition is mask_i condition val.

Normalization

EXAMPLES

Given file `85_0112.nc':

netcdf 85_0112 { dimensions: lat = 64 ; lev = 18 ; lon = 128 ; time = UNLIMITED ; // (12 currently) variables: float lat(lat) ; float lev(lev) ; float lon(lon) ; float time(time) ; float scalar_var ; float three_dmn_var(lat, lev, lon) ; float two_dmn_var(lat, lev) ; float mask(lat, lon) ; float gw(lat) ; }

Average all variables in `in.nc' over all dimensions and store results in `out.nc':

ncwa in.nc out.nc

Every variable in `in.nc' is reduced to a scalar in `out.nc' because, by default, averaging is performed over all dimensions.

Store the zonal (longitudinal) average of `in.nc' in `out.nc':

ncwa -a lon in.nc out.nc

Here the tally is simply the size of lon, or 128.

Compute the meridional (latitudinal) average, with values weighted by the corresponding element of gw (24):

ncwa -w gw -a lat in.nc out.nc

Here the tally is simply the size of lat, or 64. The sum of the Gaussian weights is 2.0.

Compute the area average over the tropical Pacific:

ncwa -w gw -a lat,lon -d lat,-20.,20. -d lon,120.,270. in.nc out.nc

Here the tally is 64 times 128 = 8192.

Compute the area average over the globe, but include only points for which ORO < 0.5 (25):

ncwa -m ORO -M 0.5 -o lt -w gw -a lat,lon in.nc out.nc

Assuming 70% of the gridpoints are maritime, then here the tally is 0.70 times 8192 = 5734.

Compute the global annual average over the maritime tropical Pacific:

ncwa -m ORO -M 0.5 -o lt -w gw -a lat,lon,time -d lat,-20.0,20.0 -d lon,120.0,270.0 in.nc out.nc

5. Contributing

Charlie Zender

Concept, design and implementation of NCO from 1995--2000

Henry Butowsky

Min, max, total, and non-linear operations. Type conversion for arithmetic. Various hacks. Migration to netCDF3 API.

Bill Kocik

Memory management

Juliana Rew

Compatibility with large PIDs

Keith Lindsay

Excellent bug reports

General Index

Jump to:	N

Footnotes

(1)

To produce these formats, `nco.texi' was simply run through the freely available programs texi2dvi, dvips, texi2html, and makeinfo. Due to a bug in TeX, the resulting Postscript file, `nco.ps', contains the Table of Contents as the final pages. Thus if you print `nco.ps', remember to insert the Table of Contents after the cover sheet before you staple the manual.

(2)

The Cygwin package is available from
http://sourceware.cygnus.com/cygwin
Currently, Cygwin 20.x comes with the GNU C/C++/Fortran compilers (gcc, g++, g77). These GNU compilers may be used to build the netCDF distribution itself.

(3)

The ldd command, if it is available on your system, will tell you where the executable is looking for each dynamically loaded library. Use, e.g., ldd `which ncea`.

(4)

The Hierarchical Data Format, or HDF, is another self-describing data format similar to, but more elaborate than, netCDF.

(5)

One must link the NCO code to the HDF4 MFHDF library instead of the usual netCDF library. However, the MFHDF library only supports netCDF 2.x calls. Thus I will try to keep this capability in NCO as long as it is not too much trouble.

(6)

The ncrename operator is an exception to this rule. See section 4.9 ncrename netCDF Renamer.

(7)

The terminology merging is reserved for an (unwritten) operator which replaces hyperslabs of a variable in one file with hyperslabs of the same variable from another file

(8)

Yes, the terminology is confusing. By all means mail me if you think of a better nomenclature. Should NCO use paste instead of append?

(9)

Currently ncea and ncrcat are symbolically linked to the ncra executable, which behaves slightly differently based on its invocation name (i.e., `argv[0]'). These three operators share the same source code, but merely have different inner loops.

(10)

The third averaging operator, ncwa, is the most sophisticated averager in NCO. However, ncwa is in a different class than ncra and ncea because it can only operate on a single file per invocation (as opposed to multiple files). On that single file, however, ncwa provides a richer set of averaging options--including weighting, masking, and broadcasting.

(11)

The exact length which exceeds the operating system internal limit for command line lengths varies from OS to OS and from shell to shell. GNU bash may not have any arbitrary fixed limits to the size of command line arguments. Many OSs cannot handle command line arguments longer than a few thousand characters. When this occurs, the ANSI C-standard argc-argv method of passing arguments from the calling shell to a C-program (i.e., an NCO operator) breaks down.

(12)

The `-n' option is a backward compatible superset of the NINTAP option from the NCAR CCM Processor.

(13)

The msrcp command must be in the user's path and located in one of the following directories: /usr/local/bin, /usr/bin, /opt/local/bin, or /usr/local/dcs/bin.

(14)

NCO averagers have a bug (TODO 121) which may cause them to behave incorrectly if the missing_value = `0.0' for a variable to be averaged. The workaround for this bug is to change missing_value to anything besides zero.

(15)

For example, the DOE ARM program often uses att_type = NC_CHAR and missing_value = `-99999.'.

(16)

32767 = 2^15-1

(17)

Operators began performing type conversions before arithmetic in NCO version 1.2, August, 2000. Previous version never performed unnecessary type conversion for arithmetic.

(18)

The actual type conversions are handled by intrinsic C-language type conversion, so the floor() function is not explicitly called, but the results are the same as if it were.

(19)

The exception is appending/altering the attributes x_op, y_op, z_op, and t_op for variables which have been averaged across space and time dimensions. This feature is scheduled for future inclusion in NCO.

(20)

The CSM convention recommends time be stored in the format time since base_time, e.g., the units attribute of time might be `days since 1992-10-8 15:15:42.5 -6:00'. A problem with this format occurs when using ncrcat to concatenate multiple files together, each with a different base_time. That is, any time values from files following the first file to be concatenated should be corrected to the base_time offset specified in the units attribute of time from the first file. The analogous problem has been fixed in ARM files (see section 3.16 ARM Conventions) and could be fixed for CSM files if there is sufficient lobbying, and if Unidata fixes the UDUNITS package to build out of the box on Linux.

(21)

This is because ncra collapses the record dimension to a size of 1 (making it a degenerate dimension), but does not remove it, while ncwa removes all dimensions it averages over. In other words, ncra changes the size but not the rank of variables, while ncwa changes both the size and the rank of variables.

(22)

Those familiar with netCDF mechanics might wish to know what is happening here: ncks does not attempt to redefine the variable in output-file to match its definition in input-file, ncks merely copies the values of the variable and its coordinate dimensions, if any, from input-file to output-file.

(23)

If lat_wgt contains Gaussian weights then the value of latitude in the output-file will be the area-weighted centroid of the hyperslab. For the example given, this is about 30 degrees.

(24)

gw stands for Gaussian weight in the NCAR climate model.

(25)

ORO stands for Orography in the NCAR climate model. ORO < 0.5 selects the gridpoints which are covered by ocean.

Table of Contents

Foreword
Summary
1. Introduction

1.1 Availability
1.2 Operating systems compatible with NCO

1.2.1 Compiling NCO for Microsoft Windows OS

1.3 Libraries
1.4 netCDF 2.x vs. 3.x
1.5 Help and Bug reports

2. Operator Strategies

2.1 NCO operator philosophy
2.2 Climate model paradigm
2.3 Temporary output files
2.4 Appending variables to a file
2.5 Addition Subtraction Multiplication and Interpolation
2.6 Averagers vs. Concatenators

2.6.1 Concatenators ncrcat and ncecat
2.6.2 Averagers ncea, ncra, and ncwa
2.6.3 Interpolator ncflint

2.7 Working with large numbers of input files
2.8 Working with large files
2.9 Approximate NCO memory requirements
2.10 Performance limitations of the operators

3. Features common to most operators

3.1 Specifying input files
3.2 Accessing files stored remotely

3.2.1 DODS

3.3 Retention of remotely retrieved files
3.4 Including/Excluding specific variables
3.5 Including/Excluding coordinate variables
3.6 C & Fortran index conventions
3.7 Hyperslabs
3.8 Wrapped coordinates
3.9 Stride
3.10 Missing values
3.11 Operation Types
3.12 Type conversion
3.13 Suppressing interactive prompts
3.14 History attribute
3.15 NCAR CSM Conventions
3.16 ARM Conventions
3.17 Operator version

4. Reference manual for all operators

4.1 ncatted netCDF Attribute Editor
4.2 ncdiff netCDF Differencer
4.3 ncea netCDF Ensemble Averager
4.4 ncecat netCDF Ensemble Concatenator
4.5 ncflint netCDF File Interpolator
4.6 ncks netCDF Kitchen Sink

Options specific to ncks

4.7 ncra netCDF Record Averager
4.8 ncrcat netCDF Record Concatenator
4.9 ncrename netCDF Renamer
4.10 ncwa netCDF Weighted Averager

Masking condition
Normalization

5. Contributing
General Index

Short Table of Contents

Foreword
Summary
1. Introduction
2. Operator Strategies
3. Features common to most operators
4. Reference manual for all operators
5. Contributing
General Index

About this document

• 1. Section One
• 1.1 Subsection One-One
• ...
• 1.2 Subsection One-Two
• 1.2.1 Subsubsection One-Two-One
• 1.2.2 Subsubsection One-Two-Two
• 1.2.3 Subsubsection One-Two-Three     <== Current Position
• 1.2.4 Subsubsection One-Two-Four
• 1.3 Subsection One-Three
• ...
• 1.4 Subsection One-Four