Because of the inherent complexity of the problem, powerful computers need to be used to run programs modeling the atmosphere-ocean system. In a typical model, the globe is divided in cubes, or cells, 500 Km on the side and several hundred meters in height. Each cell communicates with its neighbors; it can exchange radiative energy, wind, water, etc. Equations relating the interactions of these elements are solved at a particular moment in time, and the process is repeated to give projections of climate at future times. Because these programs are very complex and slow to run, the grid is ratehr coarse and simplifications in the description of the interactions are introduced.
To check that the particular model chosen is a sound one, scientists use historic data on temperatures, gas composition, etc. to predict the climate at present time. If the prediction is off mark, changes in the models are made.
Predictions vary depending on the sophistication of the model and on certain assumptions which are made.
The greatest uncertainties in these predictions come from: