Funding granted for Dr. Martin Schreiber supporting a research stay at the National Center for Atmospheric Research in Boulder, Colorado
The simulation of the atmosphere for weather forecasting was one of the first applications on supercomputers. In the last decades, computer architectures have made it possible to increase per-core performance for free with increasing processor speed. This additional computational performance also directly related to improvements in weather forecasting quality. However, this changed around 2004 and the stagnation of the per-core speed requires a fundamental rethinking on the way how time integration of partial differential equations has to be conducted.
Dr. Martin Schreiber (TUM) and Dr. Natasha Flyer (NCAR) will conduct fundamental research in this project by evaluating the quality of time integration of representative atmospheric equations using machine learning methods. Here, the important aspect is the consideration of computational efficiency, numerical properties, and impacts on future computer architectures all at the same time as it is required for weather forecasts.
This research stay is funded by the CISL Visitor Program.