From Turing’s Cathedral (pg 172-3), describing a giant lattice covering the entire surface of the Earth:

If Laplace’s mathematical intelligence were replaced by a computing machine of unlimited speed and capacity, and if the atmosphere below 100km were spanned by a computational lattice whose mesh size were less than the scale of the smallest turbulent eddy, say one millimeter… [all predictions would prove inaccurate after one month] not because of quantum indeterminacy, or even because of macroscopic errors of observation, but because the errors introduced into the smallest turbulent eddies by random fluctuations on the scale of the mean free path (ca 10-5mm at sea level), although very small initally, would grow exponentially… The error progresses from 1mm to 10km in less than one day, and from 100km to the planetary scale in a week or two.

Jule Charney and Walter Munk’s paper presented at the Conference on the Computer and the Development of Science and Learning, Institute for Advanced Study June 6 to June 8, 1972 (sadly, I couldn’t find a text for the paper online – for a full reference, see MIT’s archive).