Developping a large-scale chemistry-transport scheme for massively parallel architectures.A chemistry-transport scheme is used in weather or climate models, as well as for air quality forecasts. The goal of this thesis is to create from a scratch a model adapted to massively parallel architectures. The major contribution is to propose a new grid for the sphere, which is used to develop an efficient and scalable parallel model. The idea is really to develop a parallel framework on which several physical and chemical processes can be "plugged in". In practice, the parallelization is based on advection using an Eulerian approach on our custom grid. Therefore, the parallelization is reduced to a domain decomposition problem, which is then dealt with using MPI and a custom analytical decomposition adapted to the grid. OpenMP was also tested for more flexibility.A major part was to develop the decomposition algorithm and to exploit the grid features. Another important aspect was debugging and profiling the application on supercomputers (clusters of CPUs).HPC skills : MPI, OpenMP, batch for starting jobs, profiling and debug (valgrind, ddt, traceanalyzer).Programming skills: Fortran, Git, Doxygen, DocbookOther skills: problems solving, academic writings and presentations.

Real-time collision detection for cloth modelling in the R&D department.