I remember my first-grade teacher, Ms. Neely, telling our class that "March comes in like a lion and goes out like a lamb." I now understand that this old saw is based on the fact that winter turns to spring during the month, but I grew up outside Syracuse, N.Y., where even late March tends to be quite leonine, so it baffled me endlessly as a child.
In Washington this year, March managed to go out in a vaguely lamb-ish way. A string of warmer days at the end still left the average temperature for the month at 43.8 degrees, according to National Weather Service data, three degrees below the historical average and considerably cooler than recent years. The average temperature last March was a record-high 56.8.
Large temperature variations from year to year have significant implications, most obviously for farmers and gardeners but also for utility companies estimating energy use, city managers budgeting for snow and sports teams worrying about scheduling. Are we getting any better at predicting the weather weeks or months in advance?
Before getting to the science, it's important to recognize that there have been false starts and inflated claims in the business of long-term weather forecasting.
Consider the most famous American weather prognosticators, the Farmers' Almanac, published in Lewiston, Maine, and the Old Farmer's Almanac, produced in Dublin, N.H. The writers of these venerable books claim to use top-secret formulas, and followers of the New Hampshire version claim it is 80 percent accurate.
It's impossible, however, to fully assess the books' accuracy, because many of their predictions read like a meteorological fortune cookie: vague enough to accommodate a wide range of weather. Both publications, for example, tend to make such predictions as "sunny, cool" in four- or five-day chunks. In any given workweek, there are usually periods of sun and some temperature variation. Does that make the prediction correct?
When the almanacs risk testable predictions, they often fail. The Old Farmer's Almanac predicted that the Atlantic corridor would average 47.5 degrees for March 2013, off by nearly four degrees. That might seem like a reasonably good guess, except that anyone with access to historical averages — that is, anyone with an Internet connection — can usually get within a few degrees by sticking near the mean. The almanac's prediction for February was 29 degrees, nine degrees below the actual temperature, and the forecast for January was off by five degrees.
Critics of the almanacs are nearly as old as the almanacs themselves. A forecaster at the U.S. Weather Bureau complained about the almanacs' inaccuracy in 1905, and a Harvard professor did the same in a public address in 1926. But we simply can't let go of the dream of weather omniscience.
Generally speaking, forecasters who make predictions months in advance rely on analog techniques, which means they look for patterns in the current weather, then find similar patterns in prior years. Their predictions are based on what happened in the past. The problem is that this technique has never been shown to work particularly well. The atmosphere is a complicated place, and it's very difficult to say that a single past year, or even a combination of past years, is enough like this year to make accurate predictions.
The National Weather Service relies more on dynamic forecasting. The agency's experts observe the weather systems at work in the atmosphere, then use mathematical and physical models to calculate what will happen next. The technique is far more reliable, but it has its limits.
"We sustain higher accuracy out to two to three days in advance; then it starts dropping off faster at days six through eight," says Louis Uccellini, director of the National Weather Service.
That's why the National Weather Service makes specific weather forecasts — high and low temperature and probability of precipitation — only seven days in advance.
For extreme weather events such as hurricanes and cyclones, the agency sometimes makes longer-term predictions, based on such things as movements of the Madden-Julian Oscillation (a cycle of atmospheric weather in the Indian Ocean and the western Pacific) or the more famous El Nino weather pattern. The Weather Service also offer a 10-month forecast, but it's extremely vague, making such predictions as an above-average amount of precipitation over the course of a season.
Making specific predictions that far ahead isn't yet possible, and it may never be, according to Uccellini. He notes that three obstacles prevent scientists from making reliable forecasts even only 10 days in advance: observation systems, numerical models and computing power.
With weather satellites proliferating, there have been tremendous improvements in global data collection over the last decade or so. Computing power has also moved forward rapidly, although the ability to run computations that divide the world into small segments demands a staggering electronic infrastructure. The models are the real sticking point, but the National Weather Service is making progress by taking a sort of "poll of polls" strategy, to borrow a phrase from political scientists.
"We're now finding that if you run an ensemble of models, merging an envelope of solutions from second and third models, you can extract a more likely solution," Uccellini says.
Testing on this combined-model approach has suggested that the National Weather Service may be able to push its official forecast out to 10 days, but no decision to do that has been made. (The agency moved from a five-day forecast to a seven-day one only in 2000.)
While we may all desperately want forecasting months in advance to help us know when to plant our begonias or plan a trip to Florida, it's worth pausing to appreciate how far we've come.
"When I was a student in the late 1960s and 70s," says Uccellini, "we were just getting 24-hour forecasts of rainfall amounts and temperatures for the first time. Our goal was a 48-hour forecast. . . . Now we can predict snowstorms early enough for stores to have winter storm sales. It's amazing."