I’m a huge fan of all things “crowd” based:
America’s Best Weather Forecast
It’s on Weather Underground, and you should start using it now.
If you start at the Bay Bridge and head west along most major streets in San Francisco, you’ll eventually get to a magical land of misery known as the Sunset. The name is a joke, and perhaps even a way to trick tourists: The sun rarely visits the Sunset, not even when it sets. The primary weather element in the Sunset is fog—thick, endless, depressive clouds of it that wash up from the ocean to completely saturate the land. I lived in the Sunset for a single, terrible year. Before I moved there, I used to be one of those snobby city-dwellers who’d look down on suburbanites who couldn’t handle San Francisco’s famously capricious climate. I’d heard the Sunset’s weather wasn’t great, but hey, how bad could it be?
It was bad. Too bad for me; after our lease was up, my wife and I moved to the suburbs. Looking back, what bothered me most wasn’t the terrible climate—though I did hate it—but the vast difference between the Sunset’s weather and the weather everywhere else. Whatever meteorological patterns applied in normal parts of San Francisco didn’t seem to apply to the Sunset, which meant that forecasts for the city held no sway there. If the weatherman said it was going to be 80 and sunny, it was probably 55 and cloudy at my house.
San Francisco’s weather is an extreme example of a common phenomenon in forecasting: microclimatic variances. We like to think of the weather as being one thing, but it’s not—the weather across town is often different, usually much better, from the weather where you are. It’s this disconnect that makes us hate weather forecasters, and it’s what makes forecasters hate the weather.
Now, finally, meteorologists have figured out how to deal with this variance. Last week Weather Underground, a pioneering weather site, launched a new forecasting algorithm that takes local variations into account. The site has been working on the new system for many years, but until last week the site’s default forecast for American locations still came from the National Weather Service. But now when you type in your zip code, you’ll get a forecast from Weather Underground’s proprietary system, called BestForecast, that’s tailor-made to where you are. There’s a good chance it will give you a more accurate picture of what next week will look like, and how much better it will be for your friends who live in better parts of the city.
Why is Weather Underground’s forecast better? It relies on more data. To tell you what’s going to happen next week, meteorologists have to know what’s going on today. Over the last few decades the National Weather Service and other agencies worldwide have built up a robust system of weather stations. There are about 1,000 National Weather Service weather stations in the United States, mainly at official locations like airports. A spokesman told me there are an additional 100 locations around the country where the NWS launches weather balloons twice a day. And finally, the service relies on volunteer reports from the public—a network of 11,000 “cooperative observers” who provide their data to the service, plus 300,000 people who’ve been trained to spot severe weather events and report what they see.
Weather Underground’s system takes most of this NWS data into account, and then it adds even more. In particular, the site has assembled a huge network of constantly updating automated weather stations. These stations are owned and maintained by weather enthusiasts—people who love to track precipitation in their own backyards. They agree to share their data with Weather Underground because the site offers free archiving; you can see what your station was reporting months or years ago, easily, from anywhere. And weather enthusiasts are a generous bunch—they love the weather so much that they want to share it with everyone, and they’ve been pestering Weather Underground for years to incorporate personal weather station data into its forecasts.
The site has complied and now has 16,000 personal weather stations in its network in the United States and 8,000 more internationally. On top of that, it incorporates 26,000 extra weather stations from the Meteorological Assimilation Data Ingest System, which is run by the National Oceanic and Atmospheric Administration. (NOAA is the parent agency of the National Weather Service, but the weather service doesn’t use every MADIS station in its forecasts.) John Celenza, Weather Underground’s meteorologist and lead developer, told me that the site has the largest network of weather stations in the world. In the Bay Area, for instance, the weather service has just 18 automated weather reporting stations. Weather Underground has those 18 plus 488 more personal stations and 269 MADIS stations. That means WU has an order of magnitude more data. What’s more, many of its stations provide data in real time—as a result, Weather Underground can update its readings and forecast every two seconds.
Over the past few years, Celenza’s team worked to incorporate all this data into a better weather-prediction machine. As they developed BestForecast, they constantly checked their predictions against reality—if yesterday’s Weather Underground forecast said that it would be 72 degrees near your house tomorrow, was it right? And did the National Weather Service’s forecast do better? The company’s decision to replace the weather service with BestForecast was precipitated by its own improving results. “We just saw that we were getting really good—there were so many positives, so many places where we were doing so much better,” Celenza says.
How much better? That’s hard to quantify. Because the weather varies so much from place to place, the quality of Weather Underground’s forecasts—and its relative performance compared to NWS—depends on where you’re talking about. “It’s probably 1 or 2 Fahrenheit degrees better overall,” Celenza says when I press him to give me a ballpark figure, but he notes that that’s just an average—in some places BestForecast could be significantly better than the weather service (and, in some places, it could be worse; the weather is unpredictable, after all). But Weather Underground is transparent about its performance. If you scroll down the page on any forecast, the site will tell you how well BestForecast has performed for your area over time, and how well the Weather Service has as well. Near my old house in the Sunset,BestForecast’s recent record is rated “Excellent” (meaning it predicted the high temperature to within 1.5 degrees Celsius) while the NWS is only “Good,” meaning it was up to 1 degree worse. Near my current address in Palo Alto, Calif., BestForecast is Good while the NWS is Fair, which means it is sometimes off by 3.5 degrees Celsius. If you like the NWS forecasts more than BestForecast’s, there’s a toggle switch to get that data instead.
BestForecast has other advantages besides better accuracy, though. By incorporating so much data, Weather Undergroud can now provide longer-range forecasts—while the Weather Service only goes out to five days, Weather Underground now provides a 10-day forecast. As Weather Underground’s data network grows—they’ve added a few thousand stations in the past year alone—the site could do even better. “In 10 years, we might get to a place where we can get a pretty good 15-day forecast,” Celenza says.
Celenza—who is, naturally, a weather obsessive—says that switching over to BestForecast made him a little sad. For years, he’s considered the National Weather Service to be the world standard in forecasting. And, indeed, the Weather Service is still the nation’s preeminent forecaster of severe weather. According to a spokesman, the agency has significantly improved the lead warning time for tornadoes. In the 1980s, you might only get five minutes of warning before a tornado, but thanks to Doppler radar and storm watchers, the weather service can now warn residents 20 to 40 minutes before a storm.
For terrible weather, then, you’re still better off relying on the government. But in clear times, you’re better off trusting the crowd.