With Smartphones and Computer Models, Do We Still Need Weather Forecasters?
Yes, for entertainment. JL
Eric Berger reports in ars technica:
For TV meteorologists, even with the rise of machines, two important trump cards remain: locality and personality. As the 10pm newscast drew near one night last month, the chief meteorologist of Birmingham's ABC-affiliate began to get worked up. Balding and characteristically attired in suspenders, James Spann is one of the most recognizable and respected local TV meteorologists in the country. But he had a familiar problem. The day had been pleasant in Alabama, and more of the same temperate spring weather lay ahead—so what the heck was he going to talk about?
“I’ve got 2 minutes and 30 seconds to fill,” Spann explained. “Everyone in my audience is going to know what the weather is going to do. Except maybe my mom. She’s 85 years old. But most everybody has looked on their phone or some other device already. So what am I going to do? Am I just going to rehash everything they already know?”
Many forecasters have been asking themselves this question lately. Two technologies have converged to rapidly displace the primary function of meteorologists. First are computers that are generally better forecasters than humans. For most types of weather, numerical weather prediction has superseded human forecast methods. And secondly, thanks to the Internet and increasingly ubiquitous weather apps on mobile devices, people have continuous, immediate access to 5-day, 7-day or 10-day forecasts. As technology drives automation and machines take job after job once performed by humans, are meteorologists next in line?
The early days
Meteorology, as a science, only became possible about a century ago after a succession of physicists worked out the basics of thermodynamics in the 1700s and 1800s. American meteorologist Cleveland Abbe is credited with making the observation around 1890 that “meteorology is essentially the application of hydrodynamics and thermodynamics to the atmosphere.”
Early in the 20th century, Norwegian physicist Vilhelm Bjerknes devised a two-step process for forecasting. First he would make observations of existing weather conditions, and then he used the laws of motion to calculate how those conditions might change over time. According to Peter Lynch, who wrote about this early history for the Journal of Computational Physics, Bjerknes considered seven basic variables when understanding weather: pressure, temperature, density, humidity, and three components of atmospheric motion. He then identified seven independent equations, including the three hydrodynamic equations of motion, to solve for future conditions.
Bjerknes's brilliant ideas form the cornerstone of modern meteorology. Unfortunately, they're completely impractical. In his day there were few surface-level observations, fewer observations above the surface, and almost no information about the atmosphere over the oceans or unpopulated areas. Moreover, there existed no efficient means of collecting or sharing what few observations did exist. Even if Bjerknes had some data, there were no calculators or computers to help solve the complicated equations he found to govern atmospheric motion.
NOAA
In 1890 the American meteorologist Cleveland Abbe recognized that ‘‘meteorology is essentially the application of hydrodynamics and thermodynamics to the atmosphere."
All that began to change in the 1940s and 1950s when computer scientists like John von Neumann and meteorologists like Jule Charney worked together to apply computer science to forecasting. They ran simplified calculations on the first electronic, general-purpose computer, the ENIAC machine in Maryland. Eventually, they succeeded in making 24-hour forecasts that showed some accuracy. Alas, it also took about 24 hours to churn through those calculations, so computer models at the time were only able to keep pace with the weather.
As computer processing power has increased in the decades since, so too has the ability of physicists and meteorologists to accurately model the atmosphere. Primitive equations have given way to the ability to solve the full set of equations Bjerknes proposed, all around the world, at all levels of the atmosphere. There has been a similar revolution in data gathering for initial conditions. Whereas Bjerknes could only rely on local observations, today twice-daily balloon soundings, sensors attached to aircraft, and satellites all feed near real-time data about the entire atmosphere into the global models.
Numerical weather prediction improved dramatically, especially since the 1970s. An analysis in Nature, published in 2015, found that forecast skill in the 3- to 10- day range has increased by about one day per decade. That is, a 6-day forecast today is as accurate as a 5-day forecast a decade ago. And this improvement shows no sign of slowing down.
IBM bets on digital
In October 2015, IBM announced that as part of a $2 billion deal it would acquire most of The Weather Company. IBM planned to pair its artificial intelligence business, Watson, with the vast weather data repository of The Weather Company. The deal excluded The Weather Channel television network. For now, IBM is betting on an almost purely digital future for forecasting—not good news for someone like James Spann.
Mary Glackin, a former senior official at the US Commerce department who joined IBM in 2015 as a senior vice president, described the acquisition of The Weather Company as a “bellwether” in the field of meteorology. The deal, she said in an interview with Ars, reflects the increased skill in computer model forecasting. It also speaks to how applicable weather and climate forecasts are to corporate interests.
IBM intends to build its business around enhancing decision making, serving as a bridge between the forecast and the decision maker. “Everybody is walking around with a forecast in their pocket, on their cell phone,” Glackin said. “The Weather Company will continue to distinguish itself by having the best forecast, but really our push is ensuring effective decision making, and we’re going to get there through data analytics.”
Traditionally forecasts have been made by a meteorologist who collects the output from one or more forecast models and then distills that information into a forecast for grid points across a certain area. The traditional example would account for temperature, winds, and chance of precipitation at each of the airports in a major metro area. This process was limited by how often a forecaster could update the grids, but this job of making 5- and 7-day forecasts is now going away.
The automation of this process has been driven both by increased computing power and mobile devices. Better computer hardware has allowed forecast models to run at higher resolutions, which means that the model has more grid point outputs. So instead of a handful of points in a large city, a model may have dozens of grid points across a metro area. Weather apps have taken advantage of these higher resolution models to offer more localized forecasts for “your backyard.” Increasingly, then, a human forecaster might have some top-level input into such forecasts today, perhaps weighing the European model more heavily than the US Global Forecasting system. But this new role largely removes the human element from the forecast loop.
IBM envisions even this limited role going away. Peter Neilley, a former US government scientist who now operates The Weather Company’s global forecasting operations, told Ars this “marginal revolution” toward machines would continue in meteorology.
“The computers fundamentally are the best forecasters we have, and as a result meteorologists are becoming more interpreters of the information, helping people make more effective decisions at the consumer level, such as go to the basement now or go buy a snow shovel now,” Neilley said.
From IBM’s perspective, humans will continue to be pushed out at both the consumer and business level. Computers will learn to interpret information, too, and provide guidance on activities. The role of humans will evolve into ever more complex decisions that require multiple sources or types of information, such as when it is prudent to order the evacuation of New Orleans in the face of a Gulf of Mexico hurricane. Complex decisions will incorporate both weather forecasts (made by the National Hurricane Center, but based almost entirely on computer model guidance) and human factors, such as how many buses are available and how long historical evacuations have taken.
“There’s no doubt that the trend in the big picture is less of a pedigree on solely meteorology, and more of a pedigree on meteorology plus analytics or data sciences or other related areas,” Neilley said. “That trend is pretty steep.” That trend is also reflected in The Weather Company’s employment. In the part of the business that makes forecasts and produces them for customers, the employment ratio is about 4-to-1 non-meteorologists to meteorologists. The company prizes data analytics over forecasting skill.
In the near future, Neilley said, the company will focus less on producing the standard 5- or 7-day forecast accessible on an iPhone and more on helping consumers make decisions such as whether they should, for example, cancel the family picnic this weekend or park the car in the garage tonight. The company is moving toward allowing customers to interact directly with the data, such as signing up for an alert when the winds are going to exceed 15mph. So where does this digital future leave James Spann and his colleagues on television? Every week he and a half dozen or so other meteorologists, primarily from TV backgrounds, host an online show called WeatherBrains, which is broadcast on YouTube. It’s a free-ranging discussion about any and all weather topics. In early May I joined them to talk about the future of forecasting and address some of the questions raised by this article.
It is not as though Spann is a luddite. When he’s not on television, he lives on social media. He has 265,000 Twitter followers, 315,000 “likes” on his Facebook page and, perhaps most impressively of all, more than 20,000 followers on Google+. He has embraced social media and the Internet. “Times are clearly changing,” he said.
TV stations are struggling with weather, often a ratings winner, and the proliferation of 5- and 7-day forecasts on mobile devices. On one hand, these are great tools, he said. “It’s a beautiful thing. Joe Q. Public can look at his phone any hour of the day or night, and they don’t have to wait for some blow-dried boob to come on television so they can see a quick look at the weather.”
In addition to apps from places like The Weather Channel, AccuWeather, WeatherBug and others, in recent years local TV stations have emulated them by releasing weather apps focused on their local communities. But even while recognizing the utility of these apps for a quick fix forecast, Spann derides them as “crap apps.” That’s because they use gridded data, often from a single model forecast, and offer just a few numbers such as high and low temperatures, probability of precipitation, and maybe winds.
“That doesn’t tell you anything,” Spann said. “If you’ve got a big day coming up, a wedding, a school field trip, or whatever, that’s crap. Most of the time the forecast is inaccurate anyway, but you want to know things like what time the rain is going to start? Is it going to be severe? What is the confidence level? Will there be hail or damaging winds? Is there risk of a tornado? Flooding? These are all things you’re not going to get off a crap app.”
TV weather segments can provide this important context. In this regard the National Weather Service, which seems to be under continual budget pressure, still performs an invaluable service by providing timely weather warnings about flash floods, tornadoes, and other imminent threats. Moreover, some local weather blogs such as Capital Weather Gang have developed a large following because they not only provide a forecast, but they explain the whys and uncertainties of weather.
This is all well and good at times when forecasts are complex or call for inclement weather—thunderstorms, floods, hurricanes, tornadoes, etc. But what of the rest of the year? In most markets, there are 300 days or so a year when the weather is pretty sedate. In Raleigh, North Carolina, where Nate Johnson is a meteorologist and executive producer of the NBC affiliate WRAL, most summer days have a temperature of about 90 degrees, partly sunny skies, and a 10 percent chance of a thunderstorm. The app on your mobile device has that covered pretty well, he acknowledges.
As TV meteorologists wrestle with keeping the weather relevant in the digital age, periodically they and their news chiefs will meet with consultants and advisers about increasing the ratings of their newscasts. “Some consultants have come out and said make every day a severe weather day,” Johnson recalled. “There needs to be something in every weather cast to try and grab people’s attention and draw them in.”
That thought may sound abhorrent to a responsible meteorologist like Johnson, but the Internet has clearly ramped up hype around weather. Someone with a weather Facebook page might hype up a winter storm in Alabama, and every little tropical system gets promoted as potentially the next Hurricane Katrina.
It is something Johnson and Spann have tried to resist. “Fighting hype on social media seems to take up a lot of my time,” Spann said. “I just don’t like the trend. When you’re hyping up every time a dog pees on a bush as some severe thunderstorm, we’re just as bad as the crap apps. We can’t do that. The minute they force me to do that I’ll retire and go work at the 7/11 down here.”
The future is now
It’s not as though meteorologists haven’t seen this day coming. Back in the 1990s Harold Brooks, then and now a respected scientist and tornado expert at the National Severe Storms Laboratory in Oklahoma, had numerous discussions with colleagues about the effect of improving computer model technology on the jobs forecasters do.
Machines are simply better at repetitive tasks, assimilating a large number of observations, communicating with other machines, and carrying out large numbers of computations in a short period of time.
“The improvement in technology over the decades should allow humans to concentrate on extreme events, be they gross model errors or life and property-threatening weather,” Brooks wrote in 1995. “Technology, which initially allowed humans to make routine weather forecasts, will soon close that avenue of human endeavor, leaving open concentration on severe events. It is likely that those of us working in meteorology in the second half of the 20th century have seen a kind of human involvement in forecasting that will be open only briefly.”
That window appears to be closing. Humans today add value at the margins. We are useful during severe weather, when pattern recognition allows us to better identify a hook echo on radar and initiate warnings than a computer. We’re useful, too, in guiding emergency managers about making difficult decisions during inclement weather. Machines may eventually catch us there, too.
For TV meteorologists, even with the rise of machines, two important trump cards remain: locality and personality. Most people in Birmingham, Alabama know who James Spann is. They trust him to tell them when they need be concerned about the weather and when not to be. And in turn, Spann knows the geography and people of Alabama. He understands when a particular model forecast may not be handling forecasts for the state’s varying elevations very well, or if it cannot resolve the seabreeze in southern Alabama off the Gulf of Mexico.
So when you watch James Spann on television, if there’s something you need to be worried about, he’ll tell you. And if not? He’ll try to be entertaining. For that early May newscast, when nothing was going on locally, he talked about the tornadoes that day in Oklahoma. He shared some photos from social media. He did a little bit of local weather. He smiled that big smile. His bald head shined. And all the while Spann tried to tell a story. He did not tell viewers what time the sunrise was, nor delve deeply into the seven-day forecast.
“We have to do something that is relevant to modern viewers,” he said. “A Ron Burgundy forecast from 1979, that’s not going to cut it. Some blow dried boob coming out and saying, 'Hey everyone, sunny skies coming up for this Thursday…' No, we have to reinvent that, and to me the biggest challenge for television is that we have to reinvent that.”
As a Partner and Co-Founder of Predictiv and PredictivAsia, Jon specializes in management performance and organizational effectiveness for both domestic and international clients. He is an editor and author whose works include Invisible Advantage: How Intangilbles are Driving Business Performance. Learn more...
0 comments:
Post a Comment