A Python interface to the Fortran-written data assimilation library - PDAF

A tutorial on using pyPDAF to build a data assimilation system without parallelisation is available here: .

There are two ways of installing pyPDAF.

• The easiest approach is using conda. Currently, pyPDAF is available from conda for Windows, Linux and MacOS. The installation can be obtained via:

conda create -n pyPDAF -c yumengch -c conda-forge pyPDAF

You can start to use pyPDAF by conda activate pyPDAF.

• In HPC or cluster environment, it might not be desirable to use compilers and MPI implementation provided by conda. In this case, pyPDAF can be installed from source

git clone https://github.com/yumengch/pyPDAF.git
cd pyPDAF
git submodule update --init --recursive
pip install -v .

The pip command compiles both PDAF V2.1 and its C interface. To customise the compiler options with the local machine, it is necessary to specify the compiler, compiler options, path to the dependent libraries. In our case, the dependent library is BLAS, LAPACK, and MPI implementation.

• The installation requires Cython, mpi4py, and numpy package.
• The Fortran compiler options need to be specified in the PDAF section of setup.cfg. Note that the -fPIC compiler option is required to create a Python package. Note that these are only relevant on non-Windows machines. For Windows machines, MSVC and Intel Fortran compilers are used by default and adaptations for other compilers will need changes in CMakeLists.txt in PDAFBuild/CMakeLists.txt and pyPDAF/fortran/CMakeLists.txt.
• Options in pyPDAF section of setup.cfg requires the following options:
  • pwd is the absolute path to the pyPDAF repository directory
  • CC is the C compiler used by Cython, e.g. CC=mpicc for GNU compiler or CC=mpiicc for Intel compiler. This option is not usable in Windows as only MSVC is supported.
  • condaBuild -- ignore this option as is only relevant for conda build scenario
  • useMKL decides if you use Intel's Math Kernel Library (MKL). If True is given, MKLROOT must be specified which is the absolute path to the static MKL library
  • LAPACK_PATH and LAPACK_LIBRARY is the path to the BLAS and LAPACK directory and the linking flag respectively. They can be delimited by ,. For example, we can have LAPACK_LIBRARY=blas,lapack. Do not give -lblas as setuptools deal with the format to the linker.
  • MPI_INC_PATH, MPI_MOD_PATH, and MPI_LIB_PATH are only relevant in Windows, which is the path to .h file, .f90 file, and .lib file respectively. These paths are usually C:\Program Files (x86)\Microsoft SDKs\MPI\Include\x64, C:\Program Files (x86)\Microsoft SDKs\MPI\Include, and C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x64 respectively.

mpiexec -n 4 python -u example/main.py

Will run the model with 4 ensemble members where each member uses 1 process.

Currently, it interfaces with subroutines of PDAF-V2.1 with an example for online coupling with PDAF using a simple model based on the tutorial from PDAF. Some interface in Python changes slightly due to different ways to handling return values in Python from Fortran.

This documentation contains more details on installation, Python interface, and implementation guide on pyPDAF. The simplified interfaces in PDAF are not supported.

Yumeng Chen, Lars Nerger

pyPDAF is mainly developed and maintained by National Centre for Earth Observation and University of Reading.