Future Tense

Panasonic (Yes, Panasonic) May Have Created the World’s Best Weather Model

And it’s not sharing.

Weather data from airplanes is fed into the Panasonic model—which then helps airlines optimize their flight paths.

Panasonic

What happens when you’ve cracked the code of the atmosphere, but you still have to report to shareholders?

Last month, the weather world was a bit stunned when, in an interview with Ars Technica, Panasonic (yes, the same Panasonic that made your TV) claimed its weather forecasting model is now among the world’s best—maybe the very best. Whether you’re a weather buff or merely a patron of our atmosphere, your first question might be: Why does Panasonic have a weather model in the first place—and what happens when a big storm hits? Will it share what it knows? (OK, that’s about three questions.)

This is the start of a new era for weather forecasting. If individual companies can now match or exceed governments when it comes to modeling our planet’s atmosphere, we may be in for a very interesting next few years. (Last year, I wrote about some truly Back to the Future II–like technology that’s now allowing some companies to provide to-the-minute weather forecasts based on networks of smartphone sensors.) Since weather is one of the important variables in the global economy, better forecasts could be a boon for everyone from farmers to Little League coaches.

Traditionally, governments have been the ones to build and run weather models. First, they chop up the atmosphere into byte-sized chunks based on input from things like weather balloons and orbiting satellites, and then they run a whole series of mathematical equations to figure out the most likely future state of weather patterns—a process that’s hugely computer intensive—indeed, some of the world’s most powerful computers are used to forecast the weather. In the United States, that processing is carried out by the National Centers for Environmental Prediction, a part of the National Oceanic and Atmospheric Administration. Meteorologists then use the results (along with a lot of other information) to construct actual forecasts for actual cities. Since government weather model output is typically shared for free, that’s what most meteorologists use.

But cheaper computing power has increasingly leveled the playing field. In 2015, the Weather Channel scored a forecast coup during a major East Coast blizzard thanks in part to its own proprietary approach to analyzing the data. That’s exactly why IBM bought the Weather Channel’s parent company last year (for the data, not for the pretty faces on TV), and that’s exactly why Panasonic has been in the weather forecasting business since 2003—weather affects everything.

According to Neil Jacobs, the leader of Panasonic’s weather forecasting division, the company’s weather business is intended in part to help airlines optimize their flight routes and to provide high-resolution temperature forecasts for commodities traders anticipating natural gas or electricity prices.

He said the company’s supercomputer is “not nearly as big as what [the National Centers for Environmental Prediction] has,” but it’s almost exclusively devoted to the weather model—unlike the government’s computer, which is split among multiple tasks. Panasonic also has a bit of an advantage because it has data independently collected from commercial aircraft, which essentially replicates a weather balloon launch each time an airplane takes off or lands. Jacobs says once Panasonic expanded its data collection from airplanes to Europe and Asia, it was more computationally efficient for it to create its own model at the same time.

NOAA already buys some of the aircraft data from Panasonic to help improve government forecast models. In past emergency scenarios going back 10 years or so (defined as days when NOAA issues a call to its offices to ramp up the frequency of weather balloon launches), Panasonic has provided all of its raw data to the government at no cost.

Panasonic hasn’t yet shared its forecast output with the government, though. During last hurricane season, Jacobs says he was in email contact with scientists at the National Hurricane Center to give them qualitative updates, “but they haven’t seen any graphics or anything.” Detailed forecast output from Panasonic’s model, even if it’s just a screenshot of the projected track and future intensity of a developing hurricane, could provide the NHC with information that might sway its forecast. Jacobs says he doesn’t have a problem sharing more quantitative information with the NHC in the future, like graphics, but he is worried about the legal implications of sharing proprietary data with the government—since a public agency is obligated to make all data it acquires public, and that could eliminate Panasonic’s proprietary forecast advantage.

But! “There’s nothing saying they have to make it available to the public in real time,” Jacobs said. So, provided NHC agrees to a 48-hour delay, for example, the Panasonic model could help inform official forecasts as soon as this year. In a tough forecasting situation, such as last year’s Hurricane Joaquin, or the 2015 New York City blizzard, the Panasonic model could be an effective tie-breaker between the Global Forecast System model, run by the National Weather Service, and the European model, which is generally considered to be the world’s best. (More on how the models work here.)

But what about that claim that Panasonic is now the world’s best? Jacobs said the company is in the process of partnering with universities to provide third-party verification and is planning to launch a public website that may include time-delayed output of its model to further increase transparency.

Panasonic didn’t do all this from scratch. The company’s model is actually a modified version of the GFS model. Since that model was taxpayer-funded, anyone can use it and modify it. And Jacobs’ team has. In response to an official request for comment from NOAA, Vijay Tallapragada, a scientist at the Environmental Modeling Center of NCEP, said there are currently no plans to match some of the Panasonic model’s features, though a major upgrade is scheduled for this month, and a future upgrade in 2017 or 2018 should boost the GFS to a similar spatial resolution, about 9 kilometers.* Models with higher resolution can typically provide more detailed forecasts.

If Panasonic’s claims are true, this is an important development, but at heart it is still just an incremental increase in forecast skill. Still, the decision to keep this model private presents an interesting ethical question: If a hurricane is heading for Miami and this model knows things that others don’t, does Panasonic have a duty to share some information?

“It is at least plausible to think that Panasonic is not obligated to share its model with the scientific community (and the public more broadly), even though it would be good for the company to do so,” wrote Toby Svoboda, an ethicist at Fairfield University, in an email. “On this view, sharing the model would be ethically desirable, but doing so would go beyond what is ethically required.” That’s because sharing the model would only increase the probability of saving lives, and there’s no way of knowing which specific individuals would be helped or harmed—or, if anyone would be helped at all. At worst, Svoboda said, if Panasonic refused to share information, it could be accused of negligence, but not responsibility for harm.

Christiana Peppard, an ethicist at Fordham University, slightly disagrees. According to Peppard, “a lot depends upon the ‘culture’ of for- and non-profit ‘weather nerds’ (so called). While the profit motive suggests proprietary trade secrets should be kept … this hardly seems like a just approach in an era of global interconnections,” she wrote in an email. She reasoned that since Panasonic has already greatly benefited from publicly funded research to produce its model, “knowledge with potential to prevent cataclysmic loss of human life needs to be shared.”

Other members of the meteorological community, many of whom are eager to see a bit more detail about the Panasonic model, also expressed a desire for greater transparency.

Jason Furtado, a meteorology professor at the University of Oklahoma, says ethics aside, there’s no way Panasonic’s model will be accepted by the broader meteorological community until it’s made at least partially public, so researchers can scrutinize it. (In fairness, that’s exactly what it seems like Panasonic is planning to do.)

But most of the ethical problems with having a high-quality private weather model might come just because of the sheer lack of attention a small team of meteorologists can devote to it. Adam Grossman, of Dark Sky (purveyor of some of the aforementioned Back to the Future–esque weather-predicting technology), provided one possible scenario: “I’d be more worried about ‘less sexy’ scenarios: What if instead of a hurricane hitting Miami, it’s a risk of flash flooding in some remote part of India? It might very well be the case that they wouldn’t tell people for the simple fact that they weren’t paying attention. Open data has the advantage of a lot of eyeballs looking at it.”

In an ideal world, Panasonic would make its entire model open source—but that’s very unlikely to happen. More realistically, the National Weather Service will license it, and its output will be incorporated into our daily forecasts. Those Little League coaches will still benefit, and Panasonic will profit. If it succeeds, it could be a model for further forecast innovations by other private companies. Long-range weather forecasts tend to be unreliable—but I think it’s safe to predict that the next several years will be very interesting.

This article is part of Future Tense, a collaboration among Arizona State University, New America, and Slate. Future Tense explores the ways emerging technologies affect society, policy, and culture. To read more, follow us on Twitter and sign up for our weekly newsletter.

*Correction, May 9, 2016: This article originally misstated that an upgrade to the GFS model this month would boost its spatial resolution to 9 kilometers. The upgrade this month will not boost the model’s resolution, but an upgrade in 2017 or 2018 is planned that will. (Return.)