Load imbalance aspects in atmosphere simulations

Atmosphere simulation models usually have a regular program structure based on a regular grid decomposition of the physical space and regular computations. However, some atmosphere or weather situations show a more irregular behavior so that at the same simulation time there is more activity in some physical regions than in other regions. In this paper we investigate how the regular program structure can be adapted to more irregular weather situations to reduce the amount of computations. We study meteorological criteria to guide the computations and discuss the influence of the program change on load imbalances and the parallelization on distributed memory machines