ClimaCell claims its service, which taps into millions of wireless devices, is 60% more accurate than traditional forecasting methods.
by Douglas Heaven
On April 14 more snow fell on Chicago than had been the case in nearly 40 years. Weather services didn’t see the heavy accumulation coming: they forecast one or two inches at worst. But when the late-winter snowstorm hit, it caused widespread disruption, dumping enough snow that airlines had to cancel more than 700 flights across the city’s airports.
One airline did better than most, however. Instead of relying on the usual weather forecasts, it listened to ClimaCell—a Boston-based “weather tech” startup that claims it can predict the weather more accurately than anyone else. According to the company, its correct forecast of the severity of the Chicago snowstorm allowed the airline to better manage its schedules and minimize losses stemming from delays and diversions.
Founded in 2015, ClimaCell has spent the last few years developing the technology and business relationships that allow it to tap into millions of signals from cell phones and other wireless devices around the world. It uses the quality of those signals as a proxy for local weather conditions, such as precipitation and air quality. It also analyzes images from street cameras. It is offering subscribers a weather forecasting service that it touts as 60% more accurate than those of existing providers such as the National Oceanic and Atmospheric Administration (NOAA).
The internet of weather
The approach makes sense, in principle. Other forecasters use proxies, such as radar signals. But by using information from millions of everyday wireless devices, ClimaCell claims it has a far more fine-grained view of most of the globe than other forecasters get from the existing network of weather sensors, which range from ground-based devices to satellites. (ClimaCell taps into those, too.)
The company has now opened a new research center in Boulder, Colorado, where it is developing a new mathematical model that turns cell phone observations into weather data that can be plugged into a simulation. The more accurate your picture of the weather today, the more accurate your forecast for tomorrow.
The model can be tweaked to focus on the region, the type of weather, and the frequency of updates a subscriber wants. That would help renewable-energy companies know how much sunshine is going to hit their solar panels or how much wind will hit their turbines, for example. Better forecasting lets power providers match up supply and demand.
“There’s always a need for better forecasting,” says weather scientist Ken Mylne at the Met Office, the UK’s national weather service. “It’s impossible to do perfect forecasts, but we keep trying to narrow that gap between impossibility and perfection.”
The Met Office is also looking at new ways to measure current weather conditions. The latest version of its simulation, launched in March, uses data from aircraft radar systems, which can provide information about the temperature and humidity of the air that aircraft pass through. “It’s given a significant improvement in forecast quality,” says Mylne.
Yet making use of things like radar and wireless signals is not easy. Mylne says you can’t just put that data straight into the simulation; you have to translate your observation into the most likely weather conditions that fit it. “There is weather information in those signals, but it’s quite deeply buried,” he explains. “Exactly how you use that data is very challenging.”
Mylne thinks that what ClimaCell is doing is a good idea in principle. But he’d like to see many rigorous comparisons with other forecasters in different locations and over several months before he is convinced the technique is as accurate as ClimaCell claims.
Tim Palmer at the University of Oxford in the UK would also like to see more comparisons with other forecasters. “It’s difficult to make a clear judgment on whether they’re doing anything useful or not,” he says. “All weather services are looking for new data, and it’s quite difficult to add value. There’s already an enormous amount of information.”
A spokesperson for NOAA said the organization welcomes new techniques from the private sector but declined to comment on the specifics of ClimaCell’s approach.
In ClimaCell’s favor, Luke Peffers, who heads the startup’s research team in Boulder, has a lot of experience in measuring weather conditions. Before joining the company, he worked for the US government carrying out forensic analyses of the atmosphere to check whether nuclear test bans were being violated. He did that by looking for signs of radiation in the weather.
ClimaCell says it has also performed retrospective simulations for periods of one to 10 years that compare favorably with observations made by others. And it says it tested its model in Israel for a three-month period during heavy floods. “We did a terrific job compared to the Israel Meteorological Service’s rain gauges,” Peffers says.
As well as providing bespoke weather updates to businesses, ClimaCell is interested in collaborating with national forecasters. It is also keen to keep tapping into new sources of data. With more and more devices being connected to the internet, the number of wireless signals is increasing. As the company likes to put it, “Everything is a weather sensor.”