Code and scripts I use for inverting glacier bed stress for Kangerd, Helheim, Jakobshavn glaciers
Dependencies:
- python
- python-scipy, python-numpy
- python-gdal
- python-netCDF4
- metis
- an MPI implementation
- libscalapack-mpi1, libscalapack-mpi-dev
- libmumps-4.10.0, libmumps-dev
- elmer
To fetch all the various data, run the command
python initialize.py -d <dem source> -f <SIA fraction> -l <triangle width> -p <mesh partitions>
where <dem source> is either morlighem or cresis depending on whose DEM you wish to use, and <SIA fraction> is the fraction of the fraction of the driving stress that the bed is assumed to support at the margins. I usually used 0.5. The triangle argument specifies the diameter of the smallest triangles in the mesh, i.e. in the fastest-flowing parts of the glacier; elsewhere, the mesh will be coarsened. This script will fetch some fairly big data sets (> 2GB total).
To build the helper functions used by Elmer, the elmerf90 executable must be on your path and libelmer.so has to be on your library search path.
To run the code, execute
python run.py -g <glacier> -r <regularization> -i <iterations> -o <output file>
where <glacier> is the name of the glacier (either "helheim", "kangerd" or "jakobshavn"), <regularization> is the value of the Tikhonov regularization parameter, <iterations> is the number of iterations used in the optimization procedure and <output file> is the (optional) location of a file to store all of Elmer's diagnostic info. Only the <glacier> argument is required; the regularization parameter and iteration count default to 1.0e10 and 50 respectively if you do not provide them.
On the machines I use, an inversion with 4 cores takes 16 hours for a small mesh like Kangerd and 2 days for a bigger one like Jakobshavn at moderate resolution.