Last update: July 23, 2014
CVMix is a transparent, robust, flexible, and well-documented open source Fortran library for parameterizing ocean vertical mixing processes. It contains the following first-order closures:
- Background [time-independent] mixing, including the Bryan and Lewis (1979) parameterization
- Convective mixing (both density-based and the Brunt-Vaisala scheme)
- Double diffusion mixing, as described in Large, et al. (1994) and modified according to Danabasoglu, et al. (2006)
- The KPP boundary layer mixing scheme from Large, et al. 1994
- Shear [Richardson number-based] mixing, including both Pacanowski and Philander (1981) and the shear mixing used in Large, et al. (1994)
- Tidal mixing, which currently only offers the method described in Simmons, et al. (2004)
The library can be built on its own and linked in to a general ocean circulation model. Additionally, CVMix is packaged with a very light-weight stand-alone driver that offers the ability to run quick (typically single column) tests to ensure everything is working properly.
Community input is both welcomed and encouraged. Documentation online at
https://github.com/CVMix/CVMix-src/wiki
goes into details about how to contribute, but basically we ask three things:
- Follow the CVMix coding policies to ensure all modules use the same formatting and all documentation is consistent.
- Run the provided suite of regression tests to ensure you did not accidentally change something in another module.
- Provide a stand-alone driver for your method so that others can ensure they do not accidentally change something in your module.
INSTALLATION NOTES
CVMix can be installed using two different methods. The original uses Make and a set of Makefiles. The new method uses CMake and two CMakelists.txt files.
Building/installing using Make
The src directory contains a Makefile and a simple 'make' should be sufficient to build the standalone driver. The first time you build, the 'cvmix_setup' utility will run and prompt you for compiler and netcdf information - it will only run once, and the info is saved in bld/.CVMix_env.
To build with netcdf, run 'make netcdf'.
The default executable is $CVMix/bin/cvmix, but it can be overwritten with 'make EXE=[executable]'.
To build just the library, run the following (from src/shared):
$ make FC=[compiler]
FCFLAGS=[compiler flags]
INC_DIR=[place to save .mod files]
LIB_DIR=[place to create libcvmix.a]
OBJ_DIR=[place to save .o files]
And then use -I$(INC_DIR) -L$(LIB_DIR) -lcvmix when you build software using the CVMix library.
Building/installing using CMake
CMake can be used in GUI mode. Further information can be found on the CMake web-page. Below is provided the command line commands to configure and compile CVMix. Note that CVMix has been build on Windows using VisualStudio and the Intel Fortran compiler but NetCDF support has not been tested.
- cd $CVMix/bld/cmake_bld
- CMake promotes out of source compilation, any such directory will do.
- cmake $CVMix
- The simplest configuration - using default Fortran compiler
- cmake $CVMix -DCMAKE_Fortran_COMPILER=ifort
- Specifying a Fortran compiler
- cmake $CVMix -DCVMIX_BUILD_DRIVER=on
- Build the CVMix driver program - off by default
- cmake $CVMix -DCVMIX_BUILD_DRIVER=on CVMIX_USE_NetCDF=on
- Include support for NetCDF in the driver model(1)
- Note that this requires proper configuration of the installed NetCDF library.
- Setting NetCDF_INCLUDE and NetCDF_LIBRARIES might help.
- cmake $CVMix -DCMAKE_INSTALL_PREFIX=$CVMix/bin
- Providing an installation folder (again, any such directory will do)
Combination of the above commands is possible.
After configuration has been done compilation is as simple as:
make
and installation by:
make install
After installation the build directory can be removed.
The support for CMake builds provides sufficient infrastructure for CVMix being included in ocean models using the GIT submodule feature. This has been used in the GOTM inclusion of the CVMix mixing models as a supplement to the original turbulence models in GOTM.
1): There is unfortunately not an official NetCDF module finder in CMake.
DIRECTORY STRUCTURE
bin/ -- Default location for the cvmix executable.
bld/ -- Contains auxiliary files needed by the build system. CompileFlags.mak has default compile flags for 5 different compilers -- gfortran, pgf90, ifort, xlf90, and nagfor, as well as ftn (the Cray wrapper for pgf90). At this time, no other compilers are supported on Cray systems. cvmix_setup is a python script that saves information about what compiler to use and where netCDF is installed in .CVMix_env.
bld/obj -- Where .o and .mod files for the stand-alone drivers are stored.
doc/ -- Contains documentation. At this point, it only has a PDF of the latest protex in-code notes and a script to generate said PDF.
git_config/ -- Contains files for recommended git setup of repository. At this time, only file is git_commit_template.txt, and running
$ git config commit.template git_config/git_commit_template.txt
Will add comments to default commit log to help organize the
commit log.
include/ -- Default location for the .o and .mod files that need to be included with libcvmix.a.
inputdata/ -- The stand-alone tidal driver requires some large netCDF files to provide initial conditions and grid information. These files are downloaded the first time you run Simmons-test.sh (due to the file size, they are not kept in the same repository as the source code).
lib/ -- Default location for libcvmix.a, the library that can be linked in to an ocean model.
reg_tests/ -- Contains functional tests for all mixing methods except convective. Each test directory contains a shell script that calls functions defined in common/ to determine what settings to use for the test. All scripts have a "-h" (or "--help") flag to print usage options and a "-nc" (or "--netcdf") flag to turn on netCDF output. Note that the "-nc" is required for the tidal mixing test because it reads in netcdf data.
All directories contain one (or many) NCL script(s) to produce a
plot from the output; by default these plots are in the PDF
format.
reg_tests/Bryan-Lewis/ -- location of the Bryan-Lewis background mixing test $ ./BL_test.sh
* plot_diff_coeffs.ncl -- makes a plot with depth on the y-axis
and diffusivity (for both columns) on the x-axis.
reg_tests/double_diff/ -- location of the double diffusion test $ ./double_diff-test.sh
* plot_diff_coeffs.ncl -- makes two plots
1. ddiff-diffuse: shows the interior diffusivity for potential
temperature (normalized by the molecular diffusivity)
in the diffusive convection regime plotted against the
inverse stratification parameter.
2. ddiff-salt: shows the interior diffusivity for salinity
(normalized by its maximum) in the salt fingering regime
plotted against the stratification parameter.
Note: these plots are comparable to the two plots in Figure 4
of Large et al.
reg_tests/kpp/ -- location of the KPP mixing test $ ./kpp-test.sh
* compare_interpolation_methods.ncl -- makes three plots
1. bldepth: compares different interpolation techniques
(linear, quadratic, and cubic) when it comes to estimating
the boundary layer depth based on the depth of the mix
layer (analytically, HBL = HMIX+2)
2. bldepth_error: the error of the three different
interpolation techniques compared to the analytic solution
3. single_col22.05.pdf: the estimated boundary layer depth
using each interpolation method when HMIX=22.05 (shows
bulk Richardson number on x-axis, HBL is point where bulk
Richardson = 0.3)
* plot_bulk_Rich.ncl -- makes the KPP-bulk_Rich plot, which
shows buoyancy, zonal wind, and bulk Richardson values at
each layer center. Compare to Figure C1 of Large et al.
* plot_flux_profiles.ncl -- makes the KPP-flux_profile plot,
which shows the non-dimensional flux profiles for momentum
and scalars as a function of the stability parameter. Compare
to Figure B1 of Large et al.
reg_tests/shear-KPP/ -- location of the shear mixing test $ ./Large_test.sh
* plot_diff_coeffs.ncl -- makes a plot with the local gradient
Richardson number on the x-axis and the temperature
diffusivity coefficient (normalized by its maximum) on the
y-axis. Compare to Figure 3 of Large et al.
reg_tests/tidal-Simmons/ -- location of the tidal mixing test $ ./Simmons_test.sh -nc
* plot_diff_coeffs.ncl -- shows how the tidal diffusivity
changes over depth for a specified column (column is set in
the Simmons_nml namelist)
* plot_tidal_energy_map.ncl -- produces a global contour map
for the tidal energy flux (taken from the inputdata/
directory)
reg_tests/common/ -- a set of bash shell scripts that are used by the individual scripts mentioned above to reduce the amount of duplicate code from test to test.
src/ -- Contains the source code, organized as follows. The top directory contains modules needed by the stand-alone driver (output, for example) as well as the driver routine itself. Also contains the Makefile used to build the cvmix executable.
src/drivers/ -- Subroutines called by the driver (one per test).
src/shared/ -- Where all the modules that are needed to use CVMix with an outside model are stored. Also contains the Makefile used to build the libcvmix.a library.
templates/ -- Contains a template F90 file for adding a new mixing module to the src/shared/ directory. See the CVMix webpage for directions for requesting to have your module included in the main CVMix repository.
ABOUT THE STAND-ALONE CVMIX DRIVER
There are five options for the stand-alone driver which can be used to test two different mixing methods. The first test is to output Bryan-Lewis mixing on two columns (a high latitude column and a tropical column), and output the tracer diffusivity coefficients at each level. This tests both the Bryan-Lewis portion of the background mixing module and the ability for the CVMix data type (used for the "wrapped" interface of all of the coefficient computation routines) to use pointers or memory copies to store column data.
The second driver tests double diffusion mixing by setting up two columns, one in the salt fingering regime and one in the diffusive convective instability regime. In each case, the stratification parameter is set and then the temperature and salinity diffusivity coefficients are computed.
The third driver runs a set of 6 tests for KPP boundary layer mixing:
- Compute the boundary layer depth for a column with a given HMIX (HBLT should = |HMIX|+2)
- Compute shape function coefficients with G(0) = G(1) = G'(1) = 0 and G'(0) = 1; should be sigma*(sigma-1)^2
- Compute nondimensional flux profile over range of stability parameter values
- Compute diffusivity in the boundary layer when boundary layer is (a) below cell center and (b) above cell center
- Compute bulk Richardson number from buoyancy and zonal wind data
- Compute velocity scales (a) when surface buoyancy flux is 0 and friction is positive; and (b) when surface buoyancy flux is negative and friction velocity is 0.
The fourth driver sets up a single column using the shear mixing formula found in the Large, et al. KPP paper. For this test, each level is a different local gradient Richardson number rather than a different depth to show how tracer diffusivity varies with different Richardson numbers.
The fifth driver reads in tidal energy flux and buoyancy frequency data from an input data set and computes diffusivities in a column based on these physical values.