Why the weather forecaster’s job is harder than you think
Meteorologist’s computer experiments in 1960s led to first realization that systems could be drastically altered by tiny changes in the initial conditions
In the new book 18 Miles: The Epic Drama of our Atmosphere and its Weather, Toronto author Christopher Dewdney explores our obsession with the atmosphere.
Edward Lorenz (1917-2008) had always been a weather buff. The changeable New England weather was sure to inspire a young meteorologist like Edward, and as a boy growing up in West Hartford, Conn., he built a small weather station in his parents’ backyard, not unlike Luke Howard’s some 50 years earlier. Its centrepiece was a special thermometer that kept an automatic record of the daily high and low temperatures with little sliding markers.
He’d check the temperatures twice daily and record the numbers in a notebook. He also had a passion for math, although his twin interests seemed like two solitudes. He could measure the sensation of warmth and cold and translate those sensibilities into averages and means, but that’s as far as it went. They weren’t part of elegant, mathematical equations.
His mind craved logical challenges. On weekends and weeknights, he spent hours poring over problems in mathematical puzzle books, sometimes enlisting his father’s help. As Lorenz got older, he began to lean more toward mathematics. In fact, after graduating from Dartmouth College in1938, he went on to get a master’s of mathematics from Harvard. But then the Second World War intervened.
The Army Air Corps needed meteorologists, and Lorenz was more than qualified with his Harvard degree, not to mention that he was still a passionate weather buff. He landed a plum posting as a military weather forecaster, a job that kept him out of combat. But the stakes were high. As the war deepened, the pressure to come up with accurate forecasts was fierce — other young men’s lives were on the line. [Lewis Fry] Richardson’s equations were not yet in use, so accurate long-term
forecasting was impossible. Meteorology was still an approximate science, based as much on intuition as it was on reading instruments or the look of clouds.
While Lorenz was second-guessing forecasts in the Army Air Corps, his fellow meteorologists were more interested in theory than pragmatics. The 1940s was a period when academic meteorologists derided seat-of-the-pants forecasting. They much preferred the cleaner, more elegant theoretical side of meteorology, one in which potentially inaccurate forecasts didn’t put their reputations at risk.
But Edward gained a lot of hard experience during the war, and by its finish, he knew weather as well as any individual could. Yet he also had unfinished mathematical business. There was something about those seemingly random sequences of daily highs and lows that he had recorded as a child, something lurking behind the numbers.
Fifteen years after the war ended, Lorenz was on the faculty at one of the world’s top research facilities, the Massachusetts Institute of Technology, and in 1962 he was appointed professor of meteorology. He had become a fixture at MIT with a reputation among his peers of being a little preoccupied and distant. That must have been a feat, given how many others in the faculty shared those characteristics. On top of which, he didn’t look the part of a scientist — he had a downhome, somewhat rural look, a weathered face and piercing gaze.
It was Lorenz who first thought of using a computer to mimic the fluid dynamics of the atmosphere. In the late 1950s, at a time when small computers, especially ones that occupied less than a room, were hard to come by, the Royal typewriter company had released the Royal McBee, a “compact computer” the size of a large desk. It had a keyboard to enter programs and commands and a printer to output results. It was like a Macintosh computer decades before there were any, though it cost around $16,000, the price of a modest, two-bedroom bungalow.
Lorenz persuaded MIT to buy him one. (The rest of the faculty was skeptical about the ability of such a small computer to contribute anything meaningful to hard science, but they were all fascinated by his new toy.) He programmed it to simulate world weather patterns. In his Royal McBee was a microcosm of the world, a planet within the planet. He modelled prevailing winds, high- and low-pressure systems, temperatures — and then he let the whole thing run on its own. Every once in a while, he’d check up on what was happening by looking at printouts that translated the changing weather on this small, ideal planet into wavy lines on a graph.
One day while running the weather program, he decided to skip ahead and pick up the sequence in midstride, as it were. To set the machine at the initial conditions, he typed in the number from an earlier printout and let the Royal McBee run the equations again while he went out to grab a cup of coffee. When he returned, he knew something was wrong.
The lines on the new printout were diverging from the original, even though the number he’d typed in was identical to the first sequence. Well, almost identical. He had shortened the sequence by a mathematically infinitesimal amount; it couldn’t have produced such a divergence. Or could it? This is when Lorenz first intuited that weather systems could be completely altered by the smallest change in the initial conditions.
He rechecked his math and went to work. His revolutionary paper describing this process, delivered in 1972, was titled “Predictability: Does the Flap of a Butterfly’s Wings in Brazil Set Off a Tornado in Texas?”
It was a bombshell, exploding in the midst of the increasingly confident field of computer meteorology. The early chaos scientists called the butterfly effect “sensitive dependence on initial conditions” wherein a small perturbation at the beginning could cascade upward through the whole system, altering everything. A popular folk saying sums it up: For want of a nail the shoe was lost; For want of a shoe the horse was lost; For want of a horse the knight was lost;
For want of a knight the battle was lost;
For want of a battle the kingdom was lost!
It became a David and Goliath conflict. Lorenz’s McBee had challenged the biggest computer in the world, the Cray supercomputer at the European Centre for Medium-Range Weather Forecasts in Reading, England, which used the von NeumannRichardson algorithms. The stone in Lorenz’s slingshot was his simple simulation program that successfully, as it turned out, modelled the susceptibility of the Earth’s atmosphere to small initial changes.
If a supercomputer, with a far greater processing power than today’s forecasting supercomputers, was connected to weather sensors stationed on every square mile of the Earth’s surface and every square mile of the atmosphere and ocean depths, how accurately would it be able to predict the weather? According to [mathematician Pierre-Simon, marquis de] Laplace’s theory of the omniscient intelligence, if the machine were perfect, then it would match reality in lockstep for millennia. But according to Lorenz, and even some of today’s meteorologists, it would fall behind reality in a relatively short period of time.
Jagadish Shukla, a climatologist at George Mason University, remarks that today’s forecasting computers can forecast the weather fairly accurately five days into the future. But there’s a limit, he insists, beyond which even the most powerful computer can venture: “We may not be able to get beyond Day 15. [No] matter how many sensors you put in place, there will still be some errors in the initial conditions, and the models we use are not perfect … the limitations are not technological. They are the predictability of the system.”
Perhaps we will never, ultimately, be able to forecast much beyond a fortnight. We can know the climate; we can predict the average temperature of the ocean for decades, even centuries ahead; but we cannot know what will happen in Brazil or eventually in Texas. Excerpted from 18 Miles: The Epic Drama of Our
Atmosphere by Christopher Dewdney. © Christopher Dewdney 2018. Published by ECW Press, Ltd. Available now wherever books are sold.