HPC for Environmental Simulations

Ralf-Peter Mundani

Chair for Computation in Engineering
Technische Universität München, Arcisstraße 21, 80333 München, Germany

ABSTRACT

Due to the increasing availability of 3D geometric models on different scales, entire cities, landscapes, or even countries can be made available through digital information. On the large or global scale, Geographic Information Systems (GIS) are typically used, while on the small or local scale Building Information Models (BIM) are nowadays available for more and more buildings and infrastructure. These GIS/BIM models come with a high degree of detail including semantics or context information.  By facilitating both global and local model information in a mashup (such as Google  Earth), a multi-scale data exploration becomes possible that allows users to exploit detailed information on different scales.

One question is how these multi-resolution GIS/BIM models can be made available to multi-scale numerical simulations in order to evaluate effects of natural catastrophes such as earthquakes or floods on urban regions, built infrastructure, and also single buildings. In this presentation, we will address different aspects of the simulation pipeline such as geometric modelling, data distribution, high-performance computing, efficient parallel numerical solvers, and visual postprocessing in order to setup an HPC-based framework for (real time) environmental simulations. Therefore, we will focus on suitable distributed data structures to store the geometric models using a forest of space-trees as well as how to derive all necessary input for a numerical flow simulation on several scales (city -> block -> building -> room).
Furthermore, we will show state-of-the-art numerical methods such as multigrid in order to solve the resulting, typically huge systems of linear equations. Finally, we will present a possibility to visually explore the results in real time, i.e. during computation without exceeding any bandwidth limitations between HPC back-end and visualisation front-end.

The applicability of the HPC-based approach presented here is shown by simulating urban floods including surface flow of a city, the pipe network interaction, and its consequences to individual buildings.