Usually, the calculation of EOFs proceeds via the covariance matrix Q. For very large data sets the calculation of Q is a nearly impossible task. But it turns out that an iterative projection method exists that can generate EOFs successively without the need of calculating Q. The iterative method is computationally attractive as long as the number of EOFs to be calculated is not too large. In fact it makes something possible that is otherwise impossible.

Besides, the iterative method, while offering little advantage in terms of CPU for small data sets, is easy to understand and easy to code.

Examples will be given using the lastest Reanalysis (3/4th of million space points, 1/4th of a million time points).