Create a parallel implementation of OBLIMAP's scanning phase to enable rapid coupling of ice models with General Circulation Models (GCM) using adaptive grids
Ice caps are part of the climate system and interact with the atmosphere and ocean via various feedback mechanisms. In order to simulate this coupled system General Circulation Models (GCM's) are coupled with ice dynamical models. Our OBLIMAP-package is designed for this purpose and after its first open source release (see http://www.geosci-model-dev.net/3/13/2010/gmd-3-13-2010.html) With OBLIMAP 2.0 (see http://www.geosci-model-dev-discuss.net/gmd-2016-124/) fast embedded on-line coupling of an ice model within a GCM is now possible due to a dramatic performance gain.
OBLIMAP 2.0 is a sequential code and consists of two main phases: the scanning phase and the fast mapping phase. Despite the large performance gain for the scanning phase, it would be computational expensive if frequent coupling requires the scanning phase over and over again. For non-adaptive grids the scanning phase is only once required (as preprocessing), but for adaptive grids the scanning phase would be required for each adaptation of the grid.
By combining the current fast scanning mode with a parallel MPI implementation we expect that the time consumption of OBLIMAP's scanning phase reduces sufficiently, in order to allow on-line coupling with adaptive grids, which we expect to be a near future demand. Using more processors the fast scan method could be left out totally, and the full scan method will be parallelized using one processor for each destination grid point. The latter approach holds even for unstructered grids. The final part of the discussion (see page 17 line 15) of the manuscript which is currently published under http://www.geosci-model-dev-discuss.net/gmd-2016-124/ discusses more parallel implementation options.
The OBLIMAP 1.0 code (release 2010) is available at the Github project: https://github.com/oblimap/oblimap-1.0 The OBLIMAP 2.0 code and manual (release 2016) are available at the Github project: https://github.com/oblimap/oblimap-2.0 In this project OBLIMAP 2.0 will be subject of parallelization.
OBLIMAP uses Netcdf files, maximum size of the files is a few GB's.
OBLIMAP is written in Fortran90 and later, using netcdf. OBLIMAP compiles and runs for instance with gfortran and ifort. The netcdf files are usually visualized with ncview and by python scripts. In certain cases the preprocessing uses nco tools. We expect that the parallelization requires MPI, but for instance open-mp might be suited as well, we hope to find out such things.
See this page for help on formatting README files like this one on GitHub