The numerical simulation of micro scale interactions between environment and ground level atmosphere is a field of applied meteorology and geoinformatics that becomes increasingly important in urban planning and architecture. The multitude of different geometrical forms (buildings, plants,...) in urban areas as well as the mixture of natural and artificial materials develops a complex micro system with a high variability in space and time. Typical questions arising in this context are the thermal interaction between urban areas with high and low vegetation cover, the dispersion of air pollutants or the energetic performance of buildings.
Numerical simulations are the only possibility to comprise and understand such a dynamical and complex system and to predict is reactions towards internal (e.g. new buildings) or external changes (e.g. summer heat waves).
Numerical climate simulations are often restricted to expensive supercomputers to which most of the potential users do not have access. Contrary, the numerical models MISKAM and ENVI-met (www.envi-met.com) developed at the University of Mainz are able to analyse such complex systems by still running on a standard desktop PC.
Over decades, the development of PC-CPUs was focused on the improvement of the performance of single CPU cores. Hence, each new development cycle of CPUs was an advantage for the models profiting from the speed increase. However, during the last years the speed improvement of single CPU cores slowed down and the better performance was achieved mainly by multi core processors performing parallel tasks. This trend is furthermore supplemented by programmable graphic card processors (GPUs), which are simple in their structure, but cheap and available in huge numbers. Neither ENVI-met nor MISKAM can benefit from these additional processors or from multi-core CPUs.
The aim of the project is to develop efficient parallelisation methods for core modules of the numerical models ENVI-met and MISKAM models so that they can benefit from future CPU / GPU processingâs.
Particularly the parallelisation of the flow field calculation with the solution of the Navier-Stokes equation and the determination of geometric relations (view factor, reflections) can increase the processing speed significantly.