Algorithm for Parallel-Interactive Computing and Visualisation in Coupled Atmosphere-Land Surface Modelling

Numerical flow simulation algorithms like the ones used for Large-Eddy Atmosphere-Land Surface Modelling (LES-ALM) are based on highly irregular grid data structures. Adaptive refinement techniques are used to account for local details. The time-varying nature of flow simulation data adds an extra degree of complexity.

We adjust state-of-the-art visualisation algorithms like Direct Volume Rendering (DVR) and Texture-based Flow Visualisation to cope with large, time-dependent datasets by using out-of-core methods and parallel implementations for Multi-GPU systems and next-generation many-core coprocessors. We compare GPU-based approaches against implementations for modern many-core platforms that are typically used in HPC cluster systems. Our implementations will have to efficiently handle adaptive data structures on GPUs or many-core systems, which provide high throughput at the penalty of a restricted memory usage scenario.

The algorithms are integrated into a Virtual Reality-based Visualisation Software that is developed under the supervision of Prof. Lang.