(GRS 392) Modelling of Data Uncertainties on Hybrid Computers

Anke Schneider (ed.)

The codes d³f and r³t are well established for modelling density-driven flow and nuclide transport in the far field of repositories for hazardous material in deep geological formations. They are applicable in porous media as well as in fractured rock or mudstone, for modelling salt- and heat transport as well as a free groundwater surface.
Development of the basic framework of d³f and r³t had begun more than 20 years ago. Since that time significant advancements took place in the requirements for safety assessment as well as for computer hardware development. The period of safety assessment for a repository of high-level radioactive waste was extended to 1 million years, and the complexity of the models is steadily growing. Concurrently, the demands on accuracy increase. Additionally, model and parameter uncertainties become more and more important for an increased understanding of prediction reliability. All this leads to a growing demand for computational power that requires a considerable software speed-up. An effective way to achieve this is the use of modern, hybride computer architectures which requires basically the set-up of new data structures and a corresponding code revision but offers a potential speed-up by several orders of magnitude.