Tech start-ups are aiming for more precise predictions with new techniques, but progress is slow
In late October, the weather predictions for California’s upcoming winter weren’t looking particularly wet.
Then, in late December, the torrential rains arrived — and just kept coming. By mid-January, the state had received 200 to 600 times its usual rainfall for that time of year.
Inaccurate weather forecasts are prompting tech start-ups to rush into an industry that is dominated by government agencies and takes time to change. For example, a five-day weather forecast today has the same accuracy as a three-day forecast in the 1990s, multiple meteorologists said.
Start-ups are acquiring billions of dollars in funding aimed at getting better data for their private weather models. Private weather forecasting has been estimated to be at least a $7 billion industry. At least $880 million in venture funding has flown to weather tech start-ups in recent years, data from Crunchbase shows.
These companies are launching their own satellites and sending drones out to sea to scour for creative data points. Many are using artificial intelligence to fuel weather prediction algorithms.
Many of these start-ups aren’t quite there yet, meteorologists say. Companies make bold claims that seem unrealistic, especially when they’re often marginally better than government-provided data.
But the track they’re on is an important one, start-up executives said. As climate change fuels extreme weather events that kill people and cause billions of dollars in damage, having a better sense of when a storm is coming could save lives and lots of money.
Tomorrow.io, a weather tech company started in 2016 by Israeli military veterans, says it can make weather predictions more accurate by using proprietary forecasting algorithms, public data and “millions of different inputs” from private data to create “a much more finely tuned forecast,” according to its chief marketing officer, Dan Slagen. Private data points can include sensor data from drones and airplanes, or even when “windshield wipers go off on the car,” Slagen said.
The company provides a free app that anyone can download. But a large part of the company’s business is an intelligence platform that provides clients recommendations on how they can augment business operations depending on the forecast, according to Slagen. Airlines, for example, might be told to de-ice planes at a certain time before it gets too cold, or trucking companies might be alerted to send out their fleets earlier to beat a rainstorm.
The company’s customers, which include the U.S. Air Force, Delta Air Lines and Uber, often pay thousands to millions of dollars per year for this platform, Slagen said. And in the coming months, the company is launching the first of roughly 20 satellites because it believes space is where the best weather data lies. Having its own satellites could let the company get data quicker, Slagen said.
“[It’s] a huge game changer,” he said.
Salient Predictions, started in 2019, is trying to more accurately predict the weather a few months out. It is using machine learning, which is software that lets computers digest information and adapt on their own, to analyze global data sets with more than 20 variables, including ocean salinity, sea temperatures and pressure, wind speeds, and air temperature. The company believes its methods better predict precipitation.
Matt Stein, the company’s co-founder, said Salient Predictions’ use of machine learning puts it ahead of government agencies, which he said have “shied away” from machine learning in favor of traditional physics-based computer models. The company’s customers include Zurich Insurance, BASF and brewing company AB InBev, and the cost of its platform can start in “the low six figures,” Stein said.
And Saildrone, a company founded in 2012, has a fleet of over 100 drones resembling sailboats, many of which are powered by solar and wind power. They can be sent into the ocean for six months to a year at a time to collect data that is otherwise very hard to obtain because of the harsh nature of the sea.
Matt Womble, the director of Saildrone’s ocean data program, said making more ocean weather observations helps map storms and weather systems before they come to land. The company has been contracted by agencies such as the U.S. National Oceanic and Atmospheric Administration for missions during hurricanes. He declined to say how much it charges.
Humans have tried to predict the weather for centuries. Before modern technology, the world provided clues: “Red sky at morning, sailors take warning,” one adage goes. “Open pine, weather’s fine,” another advises, referring to pine cones opening their scales in dry air.
In the 1950s, the world saw its first computerized forecast, laying the groundwork for how weather is predicted today.
Currently, two of the world’s most prominent weather models are the Global Forecasting System, operated by NOAA and the European Centre for Medium-Range Weather Forecasts, run by an independent intergovernmental organization backed by many European countries.
In California, a drought turned to floods. Forecasters didn’t see it coming.
David Novak, the director of the Weather Prediction Center at NOAA, said in an interview that modern weather forecasting is a multistep process.
It starts with observing the atmosphere’s current state. To do that, satellites, radar and weather balloons capture basic data such as temperature, wind speeds and air moisture.
That data is fed into sophisticated models and high-powered computers, which use physics equations to create numerical representations of the atmosphere. Those are then stepped forward to predict the weather in the future, Novak added. From there, local meteorologists often analyze several models and craft forecasts for the general public.
Better satellites, more precise radar and increased computing power has made these forecasts more accurate, he said. But Novak acknowledges forecasts can be better.
How climate change will make atmospheric rivers even worse
Computer models need to better analyze the way the ocean and the atmosphere interact, he said. Satellite images need to be even better, he added. For example, satellites feed the Global Forecasting System images of the Earth in roughly 13-kilometer blocks, which Novak said should be reduced to the single digits.
“It’s an incredible amount of data points,” he said. “You need this very high-performance computing to do those kinds of fast calculations.”
It’s January, but tornadoes keep spinning up like it’s April
Some meteorologists and weather experts have concerns over the private sector getting so involved in predicting the weather.
Andrew Blum, author of the book “The Weather Machine,” said it’s worth being “very cautious” regarding start-ups’ claims that they can drastically improve forecasts, because many start-ups benefit greatly from the public data feeding their algorithms.
Blum also notes that, as extreme weather worsens, there is more incentive for start-ups to provide the best weather models, because companies want to stem the financial havoc created by storms. But that could inadvertently harm the public.
“You have a shift,” Blum said, “where people who can afford better forecasts get better forecasts.”