Why is Panasonic in the weather business? It has to do with airplanes.
The airline industry has been collecting weather data on planes for years using instruments called TAMDAR systems. In the early 2000s, AirDat was the company that designed and created these instruments. Then Panasonic acquired AirDat in 2013, and its foray into the weather industry began. Neil Jacobs, chief scientist for Panasonic Weather Solutions, spoke with Ars Technica, a technology news website.
Planes equipped with TAMDAR systems can generate vertical profiles of the atmosphere similar to weather balloon soundings — the most important input data in forecast models, Ars Technica’s Eric Berger describes:
AirDat and now Panasonic have continued to deploy these TAMDAR sensors on airplanes flying mostly regional routes over the continental United States and, more recently, Europe and parts of Asia. As the planes take off and land, they collect all sorts of useful data for forecast models, creating a vertical profile with wind, temperature, humidity, pressure, and other information. Historically, this kind of detailed data was only available from “soundings” captured by balloons sent twice daily into the atmosphere from various locations.
Panasonic is now getting 3,500 of these soundings a day, Jacobs told Ars, and that number will only increase as systems are installed on more and more planes.
More data is great for weather forecasting, and projects at NOAA have used the new data in experiments to see if it improves model forecasts. But here’s the rub: Panasonic sells this weather data to governments and organizations like NOAA, but “it found that by keeping some of the highest resolution data to itself, it could create a global forecast model that competed with the big boys,” Berger writes.
Enter the American GFS forecasting model, which is freely available to the public. AirDat and Panasonic have been tinkering with the GFS since 2008, adding the high-resolution TAMDAR data and a few significant bells and whistles. The result, Panasonic says, is a forecast model that beats not only the original GFS but also the European ECMWF, which has proved time and again to out-perform its American counterpart.
Jacobs told Ars that Panasonic measured the forecast accuracy of their model using 500 millibar anomaly correlation. In these tests, a score of “1” is perfect. Jacobs said that the Panasonic model scored a 0.926 — higher than the ECMWF’s 0.923 and 0.908 by the GFS.
If those scores are true, it’s a gigantic leap for the private weather industry. But Panasonic is not releasing much of its data, so an independent analysis of the model’s performance hasn’t been done.
Cliff Mass, a professor of atmospheric science at the University of Washington, urges caution, telling Ars that Panasonic would need to release their forecasts publicly if we’re to know how well the model is really performing. “[Jacobs has] shown me some results but not a lot of results,” Mass told Ars. “He’s making a huge claim. If it’s true, it’s extraordinary. A private company doing global data assimilation and running a global model. It would be amazing news.”
Amazing news, yes, but it raises huge questions for the weather industry, and the National Weather Service in particular. Shouldn’t the public have free access to the best possible forecast, especially considering this model might perform better in severe weather situations? Will the National Weather Service pay for it?
Jacobs shined a little light on how Panasonic is approaching the problem in his Ars interview. “Right now there’s a level playing field,” Jacobs said. “Everyone has access to the GFS and ECMWF models. But we think having exclusive access to something your competition doesn’t have, well, that’s really going to make things pretty lopsided.”
The GFS is also slated for similar improvements. In January, National Weather Service director Louis Uccellini told The Washington Post that the model would be getting two much-needed upgrades, which the Panasonic model also included in their GFS overhaul.
The first is four-dimensional variable assimilation, or 4D-Var. Basically, instead of making the assumption that all of the data points used to run the model were measured at exactly the same time, 4D-Var allows each observation to also include the time it was taken. The second addition was an ensemble Kalman filter, or EnKF, which essentially throws out bad data that would result in a poor forecast.
A spokesperson at the National Weather Service told The Washington Post that these upgrades are scheduled for May. Whether or not the improvements will push the GFS to the ECMWF’s accuracy, or Panasonic’s claimed accuracy, remains to be seen.