wanghj2015 / tempestremap

TempestRemap: Remapping software for climate applications

TempestRemap

Author: Paul Ullrich Email: [email protected]

Overview

TempestRemap is a conservative, consistent and monotone remapping package for arbitrary grid geometry with support for finite volumes and finite elements. There is still quite a bit of work to be done, but any feedback is appreciated on the software in its current form.

If you choose to use this software in your work, please cite our papers:

Paul A. Ullrich and Mark A. Taylor, 2015: Arbitrary-Order Conservative and Consistent Remapping and a Theory of Linear Maps: Part 1. Mon. Wea. Rev., 143, 2419–2440, doi: 10.1175/MWR-D-14-00343.1

Paul A. Ullrich, Darshi Devendran and Hans Johansen, 2016: Arbitrary-Order Conservative and Consistent Remapping and a Theory of Linear Maps, Part 2. Mon. Weather Rev., 144, 1529-1549, doi: 10.1175/MWR-D-15-0301.1.

Librarification for build with MOAB

The original TempestRemap sources available at https://github.com/ClimateGlobalChange/tempestremap have been modified to accomodate a robust autotools-based configuration/build system. Additional modifications include librarification of the TempestRemap sources to produce both a linkable library (enabled with --enable-linkable-library), and in the absence of such an option, the original TempestRemap drivers. These changes should be considered more of a refactor of the sources and do not contain any functional or algorithmic changes.

If you are working from the repository, please proceed with these new set of build instructions

1. cd $TEMPESTREMAP_SRCDIR && autoreconf -fi
2. mkdir -p build && cd build
3. To build a linkable library:

   ../configure --prefix=$INSTALL_DIR \ # Install dir for TempestRemap libraries
                --with-blas=$BLAS_LIB \ # Path to BLAS libraries
                --with-lapack=$LAPACK_LIB \ # Path to LAPACK libraries
                --with-netcdf=$NETCDF_DIR \ # With C++ interfaces
                --with-hdf5=$HDF5_DIR # If NetCDF was build with HDF5

4. To build the TempestRemap drivers:

   ../configure --prefix=$INSTALL_DIR \ # Install dir for TempestRemap libraries
                --with-blas=$BLAS_LIB \ # Path to BLAS libraries
                --with-lapack=$LAPACK_LIB \ # Path to LAPACK libraries
                --with-netcdf=$NETCDF_DIR \ # With C++ interfaces
                --with-hdf5=$HDF5_DIR # If NetCDF was build with HDF5

5. make all
6. make install

Additionally, a user can provide the appropriate compilers with the environmental flags (CC, CXX, FC, F77, etc.) and control the compilation/link flags (CXXFLAGS, CPPFLAGS, LDFLAGS, LIBS) as necessary.

Build Instructions with make

The software can be obtained from the GITHub repository via git:

git clone https://github.com/paullric/tempestgecore.git

You will likely need to edit the first couple lines of the Makefile to customize the NetCDF paths and change any compiler flags. Once you have modified the Makefile, build the code:

make -f Makefile.gmake all

To clean out the object file and return the sources to pristine condition, you can execute the following:

make -f Makefile.gmake clean

Mesh Generation

The remapping process requires multiple stages. First you will need an Exodus file (file extension .g) for your input mesh and your output mesh. This can either be done via the SQuadGen mesh utility, or via the three GenerateMesh executables that come with TempestRemap.

For a cubed-sphere mesh:

./GenerateCSMesh --res <Resolution> --alt --file <Output mesh filename>.g

For a latitude-longitude mesh:

./GenerateRLLMesh --lon <longitudes> --lat <latitudes> --file <Output mesh filename>.g

For a geodesic mesh:

./GenerateICOMesh --res <Resolution> --dual --file <Output mesh filename>.g

Once your input and output meshes are generated, you will need to generate the overlap mesh (that is, the mesh obtained by placing the input and output mesh overtop one another and recalculating intersections). This can be done as follows:

./GenerateOverlapMesh --a <Input mesh>.g --b <Output mesh>.g --out <Overlap mesh>.g

Offline Map Generation

Once the overlap mesh is generated, you can now generate the weight file, which the contains information on remapping from one mesh to the other. The type of offline map desired is specified by --in_type and --out_type, which can be one of the following:

fv - Finite volume mesh, with degrees of freedom stored as volume averages cgll - Continuous finite element method (such as spectral element) dgll - Discontinuous finite element method (such as discontinuous Galerkin)

Offline map generation is then performed as follows:

For finite volume to finite volume remapping:

./GenerateOfflineMap --in_mesh <Input mesh>.g --out_mesh <Output mesh>.g \
                     --ov_mesh <Overlap mesh>.g --in_np <Remapping Order> \
                     --out_map <Output map>.nc

Monotone remapping in this case can be achieved with --in_np 1.

For finite element to finite volume remapping:

./GenerateOfflineMap --in_mesh <Input mesh>.g --out_mesh <Output mesh>.g \
                     --ov_mesh <Overlap mesh>.g --in_type [cgll|dgll] \
                     --out_type fv --in_np <Input order> --out_map <Output map>.nc

Monotone remapping in this case can be achieved with argument --mono.

For finite volume to finite element remapping:

./GenerateOfflineMap --in_mesh <Input mesh>.g --out_mesh <Output mesh>.g \
                     --ov_mesh <Overlap mesh>.g --in_type fv --out_type [cgll|dgll] \
                     --in_np <Input order> --out_np <Output order> --out_map <Output map>.nc

Monotone remapping in this case requires --mono and --in_np 1.

For finite element to finite element remapping:

./GenerateOfflineMap --in_mesh <Input mesh>.g --out_mesh <Output mesh>.g \
                     --ov_mesh <Overlap mesh>.g --in_type [cgll|dgll] \
                     --out_type [cgll|dgll] --in_np <Input order> \
                     --out_np <Output order> --out_map <Output map>.nc

Monotone remapping in this case requires --in_np 1 and --out_np 1.

In each case, the linear weights file will then be written to <Output map>.nc in SCRIP format (although it’s a bare-bones version of SCRIP format at the moment and I’m not sure it’ll work with SCRIP utilities). Now that the map is generated you can apply it to your data files:

Offline Map Application

The offline map can be applied using the ApplyOfflineMap utility:

./ApplyOfflineMap --map <Output map>.nc --var <Comma-separated list of variables> \
                  --in_data <Input data>.nc --out_data <Output data>.nc

The remapped fields should then appear in <Output data>.nc. Note that if your output mesh is rectilinear, such as a latitude-longitude mesh, the data will automatically be arranged with horizontal spatial dimensions lat and lon.

Summary

Please let me know if you have any problems / bugs / comments / feature requests.