ParFlow is an open-source, modular, parallel watershed flow model. It includes fully-integrated overland flow, the ability to simulate complex topography, geology and heterogeneity and coupled land-surface processes including the land-energy budget, biogeochemistry and snow (via CLM). It is multi-platform and runs with a common I/O structure from laptop to supercomputer. ParFlow is the result of a long, multi-institutional development history and is now a collaborative effort between CSM, LLNL, UniBonn and UCB. ParFlow has been coupled to the mesoscale, meteorological code ARPS and the NCAR code WRF.
For an overview of the major features and capabilities see the following paper: Simulating coupled surface–subsurface flows with ParFlow v3.5.0: capabilities, applications, and ongoing development of an open-source, massively parallel, integrated hydrologic model.
The Parflow User Manual is available at Parflow Users Manual. The manual contains additional documentation on how to use ParFlow and setup input files. A quick start is included below.
To cite Parflow, please use the following reference.
If you use ParFlow in a publication, please cite the these papers that describe model physics:
- Ashby S.F. and R.D. Falgout, Nuclear Science and Engineering 124:145-159, 1996
- Jones, J.E. and C.S. Woodward, Advances in Water Resources 24:763-774, 2001
- Kollet, S.J. and R.M. Maxwell, Advances in Water Resources 29:945-958, 2006
- Maxwell, R.M. Advances in Water Resources 53:109-117, 2013
If you use ParFlow coupled to CLM in a publication, please also cite two additional papers that describe the coupled model physics:
- Maxwell, R.M. and N.L. Miller, Journal of Hydrometeorology 6(3):233-247, 2005
- Kollet, S.J. and R.M. Maxwell, Water Resources Research 44:W02402, 2008
The ParFlow website has additional information on the project:
You can join the Parflow mailing list to communicate with the Parflow developers and users. Join our mailing list over via:
A Parflow blog is available with notes from users on how to compile and use Parflow:
To report Parflow bugs, please use the GitHub issue tracker for Parflow:
Important note for users that have built with Autoconf, the CMake configure process is one step by default. Most builds of of ParFlow are on MPP architectures or workstations where the login node and compute nodes are same architecture the default build process builds both the ParFlow executable and tools with the same compilers and libraries in one step. This will hopefully make building easier for the majority of users. It is still possible to build the two components separately; see instruction below for building pftools and pfsimulator separately.
CMake supports builds for several operating systems and IDE tools (like Visual Studio on Windows and XCode on MacOS). The ParFlow team has not tested building on platforms other than Linux; there will likely be some issues on other platforms. The ParFlow team welcomes bug reports and patches if you attempt other builds.
Decide where to install ParFlow and associated libraries.
Set the environment variable PARFLOW_DIR to the chosen location:
For bash:
export PARFLOW_DIR=/home/snoopy/parflowFor csh and tcsh:
setenv PARFLOW_DIR /home/snoopy/parflowExtract the source files from the compressed tar file.
Obtain the release from the ParFlow GitHub web site:
https://github.com/parflow/parflow/releases
and extract the release. Here we assume you are building in new subdirectory in your home directory:
mkdir ~/parflow
cd ~/parflow
tar -xvf ../parflow.tar.gzNote the ParFlow tar file will be have a different name based on the version number.
If you are not using GNU tar or have a very old version GNU tar you will need to uncompress the file first:
mkdir ~/parflow
cd ~/parflow
gunzip ../parflow.tar.gz
tar -xvf ../parflow.tarCMake is a utility that sets up makefiles for building ParFlow. CMake allows setting of compiler to use and other options. First create a directory for the build. It is generally recommend to build outside of the source directory to make it keep things clean. For example, restarting a failed build with a separate build directory simply involves removing the build directory.
You can control build options for ParFlow using the ccmake GUI.
mkdir build
cd build
ccmake ../parflow At a minimum, you will want to set the CMAKE_INSTALL_PREFIX value to the same thing as PARFLOW_DIR was set to above. Other variables should be set as desired.
After setting a variable 'c' will configure `ParFlow. When you are completely done setting configuration options, use 'g' to generate the configuration and exit ccmake.
If you are new to CMake, the creators of CMake provide some additional ccmake usage notes here:
https://cmake.org/runningcmake/
CMake may also be configured from the command line using the cmake command. Instructions to build with different accelerator backends are found from separate documents: CUDA, OpenMP. The default will configure a sequential version of ParFlow using MPI libraries. CLM is being enabled.
mkdir build
cd build
cmake ../parflow \
-DCMAKE_INSTALL_PREFIX=${PARFLOW_DIR} \
-DPARFLOW_HAVE_CLM=ONIf TCL is not installed in the standard locations (/usr or /usr/local) you need to specify the path to the tclsh location:
-DTCL_TCLSH=${PARFLOW_TCL_DIR}/bin/tclsh8.6Building a parallel version of ParFlow requires the communications layer to use must be set. The most common option will be MPI. Here is a minimal example of an MPI build with CLM:
mkdir build
cd build
cmake ../parflow \
-DCMAKE_INSTALL_PREFIX=${PARFLOW_DIR} \
-DPARFLOW_HAVE_CLM=ON \
-DPARFLOW_AMPS_LAYER=mpi1Here is a more complex example where location of various external packages are being specified and some features are being enabled:
mkdir build
cd build
cmake ../parflow \
-DPARFLOW_AMPS_LAYER=mpi1 \
-DHYPRE_ROOT=${PARFLOW_HYPRE_DIR} \
-DHDF5_ROOT=${PARFLOW_HDF5_DIR} \
-DSILO_ROOT=${PARFLOW_SILO_DIR} \
-DCMAKE_BUILD_TYPE=Debug \
-DPARFLOW_ENABLE_TIMING=TRUE \
-DPARFLOW_HAVE_CLM=ON \
-DCMAKE_INSTALL_PREFIX=${INSTALL_DIR}Once CMake has configured and created a set of Makefiles; building is easy:
cd build
make
make installIf all went well a sample ParFlow problem can be run using:
cd parflow/test
tclsh default_single.tcl 1 1 1Note that the environment variable PAFLOW_DIR must be set for this
to work and it assumes tclsh is in your path. Make sure to use the
same TCL shell as was used in the cmake configure.
Some parallel machines do not allow launching a parallel executable from the login node; you may need to run this command in a batch file or by starting a parallel interactive session.
A version of the user manual is available at github : Parflow Users Manual
The user manual for Parflow may be built as part of the build when Latex is available on the system. Adding the -DPARFLOW_ENABLE_LATEX=TRUE option to the CMake configure will enable building of the documentation.
mkdir build
cd build
cmake ../parflow \
<other cmake options> \
-DPARFLOW_ENABLE_LATEX=TRUE \
-DCMAKE_INSTALL_PREFIX=${INSTALL_DIR}When make is run the documenation will be built and installed in ${INSTALL_DIR}/docs/user_manual.pdf.
Parflow is moving to using Doxygen for code documenation. The documentation is currently very sparse.
Adding the -DPARFLOW_ENABLE_DOXYGEN=TRUE option to the CMake configure will enable building of the code documentation. After CMake has been run the Doxygen code documenation is built with:
cd build
make doxygenHTML pages are generated in the build/docs/doxygen/html directory.
A number of packages are optional for building ParFlow. The optional
packages are enabled by PARFLOW_ENABLE_ value to be TRUE or
setting the _ROOT= value. If a package is enabled
with the using an ENABLE flag CMake will attempt to find the package
in standard locations. Explicitly setting the location using the ROOT
variable for a package automatically enables it, you don't need to
specify both values.
There are multiple ways to run MPI applications such as mpiexec, mpirun, srun, and aprun. The command used is dependent on the job submission system used. By default CMake will attempt to determine an appropriate tool; a process that does not always yield the correct result.
There are several ways to modify the CMake guess on how applications should be run. At configure time you may overwride the MPI launcher using:
-DMPIEXEC="<launcher-name>"
-DMPIEXEC_NUMPROC_FLAG="<flag used to set number of tasks>"An example for mpiexec is -DMPIEXEC="mpiexec" -DMPIEXEC_NUMPROC_FLAG="-n".
The ParFlow script to run MPI applications will also include options specified in the environment variable PARFLOW_MPIEXEC_EXTRA_FLAGS on the MPI execution command line. For example when running with OpenMPI on a single workstation the following will enable running more MPI tasks than cores and disable the busy loop waiting to improve performance:
export PARFLOW_MPIEXEC_EXTRA_FLAGS="--mca mpi_yield_when_idle 1 --oversubscribe"Last the TCL script can explicity set the command to invoke for running ParFlow. This is done by setting the Process.Command key in the input database. For example to use the mpiexec command and control the cpu set used the following command string can be used:
pfset Process.Command "mpiexec -cpu-set 1 -n %d parflow %s"The '%d' will be replaced with the number of processes (computed using the Process.Topology values : P * Q * R) and the '%s' will be replaced by the name supplied to the pfrun command for the input database name. The following shows how the default_single.tcl script could be modified to use the custom command string:
pfset Process.Command "mpiexec -cpu-set 1 -n %d parflow %s"
pfrun default_single
pfundist default_singleThis section is for advanced users runing on heterogenous HPC architectures.
ParFlow is composed of two main components that maybe configured and built separately. Some HPC platforms are heterogeneous with the login node being different than the compute nodes. The ParFlow system has an executable for the simulator which needs to run on the compute nodes and a set of TCL libraries used for problem setup that can be run on the login node for problem setup.
The CMake variables PARFLOW_ENABLE_SIMULATOR and PARFLOW_ENABLE_TOOLS
control which component is configured. By default both are TRUE. To
build separately use two build directories and run cmake in each to
build the simulator and tools components separately. By specifying
different compilers and options for each, one can target different
architectures for each component.
ParFlow includes a Docker file for configuring a Docker image for running ParFlow.
A Docker image for ParFlow is available on Docker hub. See the following section for how to run the Docker image. The Docker latest image is automatically downloaded by Docker when run.
The https://github.com/parflow/docker repository contains an example setup for running ParFlow in a Docker instance. See the README.md file in this repository for more information.
If you want to build a Docker image, the build script in the bin directory will build an image using the latest ParFlow source in the master branch. If you want to build a different version of ParFlow you will need to modify the 'Dockerfile' file.
./bin/docker-build.sh.\bin\docker-build.batCopyright (c) 1995-2019, Lawrence Livermore National Security LLC.
Produced at the Lawrence Livermore National Laboratory.
Written by the Parflow Team (see the CONTRIBUTORS file)
CODE-OCEC-08-103. All rights reserved.
Parflow is released under the GNU General Public License version 2.1
For details and restrictions, please read the LICENSE.txt file.