BeTR

BeTR is a standalone reactive transport libary designed to be integrated into land surface models such as CLM and ALM.

Jinyun Tang, [email protected]

References

Tang, J. Y., Riley, W. J., Koven, C. D., and Subin, Z. M.: CLM4-BeTR, a generic biogeochemical transport and reaction module for CLM4: model development, evaluation, and application, Geosci. Model Dev., 6, 127-140, doi:10.5194/gmd-6-127-2013, 2013.

Building

Requirements

Requirements for configuring and building BeTR:

• cmake > 3.1

• fortran compiler

  • gfortran > 5.3.0. Older versions make work, but are not supported. At a minimum, you will have to change the compiler flags. The pfunit testing frame work can NOT be build with gcc < 4.9.

  • intel - TBD

  • pgi - TBD

  • nag - TBD

• c compiler - only used for 3rd-party libraries. Any modern compiler should work. The following versions are tested:

  • clang - approximately 7.3

  • gcc - 5.3

• python 2.7

All third party dependancies for building and running standalone BeTR are included in the 3rd-party directory.

Build

BeTR uses a cmake based build system. The default build is debug. To build using the default debug configuration:

cd ${SBETR_ROOT}
make config
make all

This will do an out of source build in:

${SBETR_ROOT}/build/Xyz-Debug/src

where Xyz is the build configuration.

The standalone sbetr executable is in:

${SBETR_(ROOT}/build/Xyz-Debug/src/driver/sbetr

To build a release configuration of the code:

cd ${SBETR_ROOT}
make debug=0 config
make debug=0 all

The following command will create local/bin/, where sbetr will be installed.

make install

Please note, the top level make file providing make config etc is a convenience for the most common use cases. You don't have to use it and can specify all configuration and build commands manually.

HPC machines

To run betr on cluster, one needs to load the following (take intel compiler for example) the compiler, cmake, python (>2.7), and mkl, the configuration command is then

make config CC=icc CXX=icpc FC=ifort

and the install command is

make install CC=icc CXX=icpc FC=ifort

Others should be done similarly as one run betr on a desktop or laptop, with appropriate modifications as described above.

• yellowstone

  • intel

    module unload ncarbinlibs
    module load intel/16.0.2
    module load mkl/11.3.0
    module load cmake/3.3.1
    module load python/2.7.7
    
    make CC=icc CXX=icpc FC=ifort config all install test

  • gnu - doesn't work yet, can't link blas

    module swap intel gnu/5.3.0
    module load cmake/3.3.1
    module load python/2.7.7
    
    make config all install test

  • pgi - in progress

    module swap intel pgi

• edison

  • intel

    module load cmake/3.3.2
    module load intel/15.0.1.133
    
    make CC=icc CXX=icpc FC=ifort config all install test
    

• cori

  • intel

    module load cmake/3.3.2
    module load intel/16.0.0.109
    
    make CC=icc CXX=icpc FC=ifort config all install test
    

Testing

BeTR testing inclueds pFUnit based unit tests and a systems level regression test driver.

pFUnit tests are created automatical during the build. Run the unit tests with

make test

or by calling ctest in the build directory.

Regression tests are based on calling the standalone sbetr executable and checking the results are within a specified epsilon of a baseline.

Regression testing will eventually be integrated into the 'make test' command with unit tests. For now they have to be run separately.

cd regression-tests
make rtest

Creating new tests

Regression tests

Regression tests are grouped into test suites, which are defined by configuration files. The directory where the configuration file is referred to as the 'suite directory'. A suite directory can contain multiple configuratin files. Configuration files are in cfg/ini format, and have the following information:

[default_tolerances]
#category = value type
general = 1.0e-14 absolute
concentration = 1.0e-13 relative

[mock]
# override default tolarance
concentration = 1.0e-14 absolute
timeout = 30 seconds

Where there is one required section: default_tolerances. This contains the default tolerances for all tests in this suite. Tolerances are specified by category, concentration, velocity, general. The type of toleraces can be absolute, relative or percent. These values can be over ridden for individual tests.

In general, tests should be kept short. Long tests are harder to debug and they rarely provide a benefit over shorter tests. If a particular set of conditions needs to be tested, it is better to create a special forcing data set that exercises those conditions in a short test. The test suite enforces a timeout limit for tests to prevent infinite loops and other hangs from running. If a test is exceeding the default timeout limit, then it can be increased on a case by case basis in the test suite configuration.

All other sections are considered to define tests. The name of the section is the test name. It is expected that a test_name.namelist file will be present in the same directory as the suite configuration file. sbetr is run from the same directory as the configuration file, and all paths in the namelist file must be relative to this directory. sbetr will write a test_name.regression file with the regression test data. This is compared to test_name.regression.baseline, also in the same directory as the suite configuration file. Keys in the test section are used to modify the test, either by changing the tolerance, or modifying some other functionality, e.g. we will eventually have restart tests triggered by a setting in the test section.

Setting tolerances is a balancing act that requires some trial and error. Test tolerances should be set as tightly as possible to identify small changes in behavior. But they should be loose enough to be platform independent, so we don't get false positives when moving to a new platform.

A requirement for testing is that the repository be self contained and platform independent. All input data and baselines should be contained in the repository. Inorder to be revision control friendly, the data should be saved in the repository as plain text files. Netcdf files should be converted to their text representation with:

ncdump -p 9,17 ${DATA}.nc > ${DATA}.nc.cdl

These files will be automatically converted back to binary when the test suite is run with:

ncgen -o ${DATA}.nc ${DATA}.nc.cdl

Check the cdl file to ensure all variables have the same data type as is expected in the code, i.e. double percision for r8. Before saving a cdl file in the repo, verify that the round trip of cdl-nc-cdl results in the same files. Failure to do so will result in unreproducible test results!

Unit tests

Document procedure for new pFUnit tests

Running

sbetr takes exactly one command line arguement, the name of the input namelist file. The namelist input file specifies runtime configuration and paths to other input data files. NOTE: paths are relative to the directory where sbetr is executed!

cd ${SBETR_ROOT}/example_input
${SBETR_ROOT}/local/bin/sbetr mock.namelist

The example is set with mock run that transport five tracers: N2, O2, AR, CO2, CH4 and DOC

Development

Key direcotries:

• 3rd-party - select 3rd-party sources that betr depends on.

• src - contains model code

  • src/betr - LM independent betr library

  • src/driver - standalone driver

  • src/stub_clm, esmf_wrf_timemgr, src/shr - stub version of land model code needed to make the standalone LM interfaces compile.

  • Application - place to hold customized betr applications

• cmake - contains utilities for the configuration and build system.

• regression_tests - regressiont test input and baselines.

To configure a new bgc implementation, follow the example in BGCReactionsMockRunType and add the new configuration name to ApplicationsFactoryMod.