Box scores tell you how a baseball team has done. Stock listings tell you how a stock has performed. But I've never seen a mechanism indicating whether weathermen have any idea what they're talking about. Does anyone keep track of how accurate they are? -- Steven Goldberg
Is there any evidence that computer tracking has improved the accuracy of weather forecasting? Despite the introduction of cutting-edge technologies, the weatherpeople seem to get it wrong as much as ever.
Illustration by Slug Signorino
Now, Tom, be fair. Weather forecasting is one of those things, like hairpieces and housework, that attract notice only when there’s a problem. I’m guessing you’ve never had a coworker turn to you and gush, “They said the high would be 78 yesterday, and you know what it was? Seventy-eight! Man, that National Weather Service is something else!”
As you’d imagine, advances in weather prediction closely follow advances in technology and communications. Early forecasters had to make their best guesses using only basic gear (thermometer, barometer, etc) and personal experience with local conditions. In 1743 Benjamin Franklin (who else?) compiled reports from colonial postmasters to track a hurricane’s progress up the eastern seaboard. By 1848 weather dispatches were traveling via telegraph, and in 1871 the newly founded U.S. Weather Bureau started publishing the first general forecasts three times a day. Weather balloons carrying radiosondes went up in the 1930s, providing a look at doings in the upper atmosphere. Finally, mathematical weather-system models, first proposed in the early 1900s, came into their own circa midcentury, when (a) computers got powerful enough to handle the calculations needed to simulate atmospheric movement and (b) radar and weather satellites greatly increased the available data.
And as computer models get updated (to include things like ocean and land effects and chaos theory) and more weather stations get sampled, forecasts do, in fact, get better. According to the American Meteorological Society, sea-level pressure forecast accuracy doubled between 1977 and 1987, and by 1991 a five-day weather forecast was as good as a three-day forecast from 1981. One study of extended forecasts from 1997 to 2004 found that using a new projection system improved temperature predictions by as much as half a degree Fahrenheit, while four-day rain forecasts improved by as much as 8 percent.
One key function of meteorology is predicting violent weather, particularly thunderstorms that could produce hail, flooding, or tornadoes, and here too there have been real advances. From 1973 to 1996 the chance of predicting a severe storm increased from 30 percent to 66 percent; between ’78 and ’96 the odds of a serious tornado being anticipated by a watch announcement rose from 48 percent to 95 percent. In ’91 the average tornado lead time was six minutes; by 2004, thanks to NEXRAD Doppler radar, it was up to 13. Between 1990 and 2006 the National Hurricane Center greatly bettered its path projections for tropical cyclones in the Atlantic – for one-day forecasts the average tracking error dropped from about 124 miles to about 58, and from over 350 miles to less than 175 for three-day forecasts. (Interestingly, though, over the same period predictions of the storms’ intensity didn’t improve at all.)
Of course, forecasting’s a lot easier when the weather doesn’t change much. In 2005 ForecastWatch.com studied online temperature forecasts for 689 U.S. cities and found, unsurprisingly, that they were most reliable in the south and the southwest, on the west coast, and in Hawaii. Frankly, if you can’t predict the weather in Honolulu, where the high’s in the 80s and the low’s around 70 nearly all year, you should look up “anterograde amnesia” right now, before you forget. Conversely, temperatures were hardest to foresee in the Great Plains and the northeast, where the range of possibilities is much greater. Likewise, winter forecasts are far less reliable than summer ones.
Science, however, continues to make headway. Infrared satellite scans have recently increased six-day forecast accuracy by 4 percent; the new high-resolution Weather Research and Forecasting model can reportedly cut errors in nighttime temperature and humidity prediction in half and can also help planes avoid turbulence.
Now: is anyone keeping tabs on how specific forecasters do? Sure. The aforementioned ForecastWatch compares projections by major national forecast providers and makes zip-code-specific rankings available free online. (Founder Eric Floehr won’t say whether one provider is superior overall, claiming his findings can’t fairly be boiled down that far.) A Phoenix firm called WeatheRate goes further: claiming to be the nation’s only independent verification service for TV weather forecasts, it reviews four-day projections from local stations in about 75 cities year-round and determines a leader in each market. I don’t know if that pile of underexploited stats is giving you any ideas, Steve, but let’s just say that when millions of meteorology fans start fielding lineups of weathermen in your online Fantasy Weather Forecast League, I’ll expect to see a cut of the sign-up fees.
Send questions to Cecil via firstname.lastname@example.org.