Atmosphere and Ocean Circulation Simulation on Massively Parallel Computers

In this paper we present some results on the implementation of atmosphere and ocean circulation models on massively parallel computers. These circulation models are enormous computationally demanding tasks and play an indispensable role in climate modelling. Therefore they have an important impact on society. After a brief introduction on the physics behind the circulation models based on geo-physical uid dynamics, the diierences between the simulation of atmosphere circulation and ocean circulation are outlined. Next we discuss the numerical techniques applied in the models to solve the partial diierential equations. Performance results of diierent implementations of a simple, theoretical model on parallel architectures are presented, as well as preliminary results for the implementation of a very computational intensive task in a numerical weather forecasting model. Finally, ideas are outlined on the development of a distributed library for the type of calculations in the models concerned. 1 General circulation models Atmospheric and ocean circulation models describe the circulation of heat, air/water and chemicals in the atmosphere and world oceans or smaller basins. In coupled models also the exchange between oceans and the atmosphere is taken into account. These circulations have a major impact on climate and climate change. The strong relation to the behaviour of climate means that a computer model designed to predict climate changes, must include the ocean circulation and mixing, as well as the atmospheric circulation and physics of clouds, radiation and precipitation. The atmospheric component is relatively well advanced, because of the economical importance of numerical weather prediction, which started in the 1950s. The rst numerical ocean models were developed in the 1970s, so this research area is relatively new, and as a result lays behind the atmospheric component. Ocean Circulation Modelling and Climate Modelling are considered as two of thèGrand Challenges' in computational sciences, as is outlined in a report by the Executive OOce of the President of the USA 1], and in more detail in a report by the Committee on Physical, Mathematical and Engineering Sciences of the National Science Foundation 2]. In Europe the importance of these research areas has been recognized by the Commission of the European Communities in 3].