The impossibility of weather forecasting

The un-predicted, extra-tropical
Hurricane Sean, from 2011

Note: this is re-printed from the March/April 2012 issue of Yacht Essentials Magazine. Thanks to Chris Kennan and Brad Kovach for permission!

“This is complicated.” That is what scientists in the 1960s and 1970s decided a simple graph depicting a chaotic curve – the ‘Lorenz Attractor’ – was trying to say, without having to speak a word.

Fifty years ago, long-range weather forecasting was already a scientific impossibility, and Edward Lorenz proved it. In 1962, Lorenz published his definitive work on meteorology in Volume 20 of the Journal of the Atmospheric Sciences, a paper made public to little fanfare at the time, but which would soon become one of the most oft-cited works in the greater science community. No one knew it then, but in hindsight, the paper was perhaps the first published example of what would become chaos theory, the simple idea that the way nature works is not so simple at all.

‘Chaos theory’ went mainstream with the publishing of Michael Crichton’s novel and subsequent release of the movie Jurassic Park in 1993. In it, Dr. Ian Malcolm, a mathematician specializing in chaos theory (played by Jeff Goldblum), is troubled by the notion of predicting and controlling nature.

Goldblum’s character rambles on about chaos theory and it’s implications in a number of fields (including weather) in the helicopter as the team descends on Jurassic Park (in real life, Cocos Island, off the Pacific coast of Costa Rica, and reknowned for its shark diving). Everyone knows the ending to the story.

“All the money spent on long-range forecasting – about half a billion dollars in the last few decades – is money wasted,” Goldblum’s character says. “It’s a fools errand. It’s as pointless as trying to turn lead into gold.”

--

Before Lorenz (and Goldblum), meteorologists were strangely confident in their ability to one day not only predict but actually control the weather, thanks mostly to the advance of the computer and the space satellite. By the 1980s, meteorologists were in fact producing fairly accurate short-term forecasting. The best came out of Reading, England, from the European Centre for Medium Range Weather Forecasts, using Cray supercomputer models. However, beyond two and three-day forecasts, the science became speculation. Beyond five and six days? The forecasts were worthless.

Sound familiar? Thirty years later, and weather forecasting has not advanced at all. Even today forecasters cannot predict beyond a couple days with any accuracy. Think of the last hurricane track forecast you have seen – the cone-shaped “possibilities” of the hurricane’s future position covers hundreds of miles after a day or two, and even then it is sometimes off (remember “wrong-way” Lenny in the Caribbean in 1999?). Lorenz offered an explanation for meteorologist’s confidence.

“There are real physical phenomena for which one can do an excellent job of forecasting,” Lorenz said, “such as eclipses and oceanic tides.”

Weather is vastly more complicated. The logic behind Lorenz presumption that the weather is inherently unpredictable lays in a recently familiar concept (thanks again to Hollywood and Ashton Kutcher) – the “butterfly effect” – scientifically known as ‘sensitive dependence on initial conditions,’ and the basis of chaos theory.

Lorenz was at heart a mathematician, and set out to model the fluid dynamics of the atmosphere in a simple math equation. Using primitive computers to perform the calculations, Lorenz discovered something incredible in his numbers, and by accident. By merely changing the decimal point in certain ‘input parameters’ – say for example, starting humidity at a given point in the atmosphere – the resulting graph that showed the ups and downs of the resulting ‘weather’ gradually drifted away from the previous graph, when in theory it should have remained the same (the input parameters were for practical purposes identical – certainly more accurate than could actually be measured). The mathematical formula was a comparatively simple one – no one expected the result. Lorenz had stumbled onto chaos theory.

The implications for weather forecasting were immediately obvious. While computers could even by then accurately model a (simple) dynamic weather system, it was (and is) impossible to exactly measure the real atmosphere needed to make accurate predictions. Weather stations are miles apart – sensors can obtain data where they are located, but can never account for the infinite variation between them. Even if sensors could theoretically measure every condition of the atmosphere at points inches apart and to the nth decimal place, the slight variations between those inches and the sheer number of variables that create weather – humidity, pressure, temperature, etc. – create the chaos. Those inconsistencies multiply in the dynamic system over time, creating more and bigger inconsistencies and the model quickly breaks down after the first few days.

In practice, modern GRIB forecasts display a wonderful picture of chaos theory in action. When GRIBs are downloaded, they offer a picture of the weather over a large scale in real-time. A simple animation can be run showing how the weather changes going into the future, based on computer models. A simple test on the accuracy of a forecast – and a perfect visual analogy to what Lorenz first discovered in 1962 – is to simply ‘animate’ the GRIB file seven days out, then download the real-time file seven days later. Do they match? Not likely.