Climate , the Ocean , and Parallel Computing

The climate of the earth is controlled by an interplay among many competing physical processes operating in the atmo-sphere and the ocean and on land. Of the many questions facing today’s climatologists, two seem particularly urgent. Is the balance among those processes being affected by human activities? And if so, how large are the resulting climatic changes relative to natural climatic variations? The high priority accorded internationally to answering those questions is impelled in part by the growing list of troubling environmental problems such as greenhouse warming, ozone depletion, pollution of the atmosphere and ocean, and tropical deforestation. Past observations of the atmo-sphere and the ocean have contributed greatly to our knowledge of the climate system but still constitute only a short and incomplete baseline of data with which future changes can be compared. More extensive observations are in progress or being planned: NASA’s Earth Observing System satellite will produce the most comprehensive picture to date of the present state of the earth’s climate; the Tropical-Ocean Global-Atmosphere Program is investigating the impact of El Niño and the Southern Oscillation on weather patterns in mid-latitude regions; and the decade-long international World Ocean Circulation Experiment will probe to great depths the circulation in all the major ocean basins. Such projects are an essential aspect of climate studies, but a theoretical framework is also needed as a basis for interpreting the accumulated data. The computer models known as general-circulation models (GCMs) provide such a basis by simulating the temporal evolution of the atmosphere or the ocean in three dimensions (latitude, longitude, and altitude or depth). In addition, GCMs are indispensable tools for investigating parts of the climate system, such as the deep ocean, that are very difficult to observe and for estimating the effects of natural and human-induced environmental changes on climate. Three-dimensional GCMs were first developed in the 1960s. Their fidelity has since been greatly increased but is still limited by a shortage of data for validating the models and by the capabilities of the computers on which they are implemented.