ALL meteorologists were optimists in the 1950s. With the advent of electronic computers and the development of mathematical models of the atmosphere, it seemed only a matter of time before every whim, every eccentricity, of the elements could be confidently anticipated. Given computers big enough to do the calculations, accurate forecasts for weeks or even months ahead would be attainable.
But as a wise man once asked: How do you make God laugh? And he gave the answer: Tell Him your plans." So it was with meteorology. As Joseph Drinan pointed out in the letters page last Friday, the Butterfly Effect ensured that the meteorologists' dream was one that could never be realised.
The Butterfly Effect is the popular exposition of "chaos theory", which in a meteorological context describes the inherent unpredictability of the atmosphere when we try to look too far ahead. Chaos theory is not confined to the weather, having applications in many other disciplines, but the concept was largely based on the work of American meteorologist Edward N. Lorenz, of the Massachusetts Institute of Technology in the early 1960s. In a famous lecture he coined the phrase: "Could the flap of a butterfly's wings over Brazil spawn the next tornado up in Texas?" In the intervening years the chaos theory has spawned a whole new glosary of jargon, from "phase spaces" with "sources", "sinks" and "saddles", to "strange attractors", "bifurcations", "tangles" and "period doubling cascades".
Chaos stirs up trouble in weather models because they cannot start out with an exact picture of the current weather situation. Their initial snapshot is based on perhaps 10,000 observations from around the world which may seem a lot, but it can only give a very rough approximation of the varying conditions over an area of some 200 million square miles.
As the computer model begins projecting into the future, the Butterfly Effect magnifies its initial imperfections. At first the error remains small and the model provides a forecast of reasonable accuracy, but with time the errors compound, growing so large that the simulation lose all reliability. Eventually, a prediction for perhaps 10 days ahead may bear no more than a chance resemblance to what actually will happen.
But within these chaotic limitations, Lorenz remains an optimist. A month or two ago he predicted that "we may some day forecast a week in advance as well as we now do for three days, and two weeks in advance almost as well as we can now forecast for a week ahead."