(Bloomberg) -- In late August and early September of last year, Hurricane Ida tore across the U.S., leaving a trail of destruction. The ferocious storm was at its most ruthless when it landed, and again as it headed back out to sea. The Louisiana coastline and the New York metropolitan area were among the hardest hit.

Hoboken, New Jersey—sitting just across the Hudson River from Manhattan—saw 70% of its streets underwater. Mayor Ravi Bhalla called the maelstrom a “manmade disaster” that was turbocharged by global warming. Ida had been the second monster storm to swamp his city in barely a week, and the fourth in only a year.

For Caleb Stratton, who as Hoboken’s chief resilience officer is responsible for preparing the city’s infrastructure for such calamities, this was the last straw. Helped by federal grants, Stratton has spent more than a decade implementing various flood mitigation measures across Hoboken. But the increasing number, volatility and unpredictability of storms has left the city gravely exposed, he said. A more proactive approach would be needed.  

“Hoboken is an older city,” Stratton said of the 173-year-old birthplace of Frank Sinatra and site of the first organized baseball game. “And these cities weren’t constructed to deal with [these] types of storms.”

His solution, while perhaps more mundane than massive infrastructure projects aimed at climate adaptation, could signal a broader, lucrative revolution in an arena long the exclusive province of the government: weather forecasting.

For decades, the small group of companies that once comprised the private weather industry would repackage data from the U.S. National Oceanic and Atmospheric Administration (NOAA) and its predecessor agencies to sell it to customers ranging from newspapers and radio stations to airlines and theme parks. But driven largely by the catastrophic effects of global warming, the industry has been expanding rapidly.

Since 2017, there have been 86 weather-related disasters in the U.S. exceeding $1 billion in damage, a much faster pace than in previous years. Twenty of those events were in 2021 alone. Indeed, climate disasters last year cost the American economy $145 billion, a stunning amount which, when accounting for inflation, is the third-costliest year on record.

While most private weather platforms still rely heavily on data from NOAA, they’re increasingly supplementing it with their own—whether from intergovernmental agencies or private sources. The Weather Company, which was bought by IBM for $2.3 billion in 2016, reportedly accumulates up to 80 million observations each day through barometers in smartphones. Additionally, cheaper access to low Earth orbit is enabling some companies to launch their own satellites—and in turn provide data to NOAA. 

This accelerated privatization of the weather comes as NOAA’s once stellar public profile was arguably dented by Donald Trump. Not only did the Republican famously display a map of Hurricane Dorian’s path featuring an extra loop drawn over Alabama (which he wrongly predicted would be hit), but NOAA officials backed up his fake forecast.

Read More: How AI Can Make Weather Forecasting Better and Cheaper

Private weather isn’t new. In 2013, an article from the Wharton School estimated that a global weather forecasting industry of around 350 companies was pulling in around about $3 billion annually. In 2017, NOAA estimated that the sector in the U.S., which today encompasses everything from hourly road surface forecasts for long-haul truckers to drought forecasts for individual farms, was worth $7 billion. Reports suggested it was growing at a rate of around 10-15% annually.

And while the industry couldn’t exist (at least not right now) without U.S. government data, NOAA is becoming increasingly reliant on the industry. NOAA spokesperson Susan Buchanan said that while the agency’s work is the foundation for private weather, it needs these new sources to create a “weather-ready nation,” since NOAA alone can’t address “all the hyper-local vulnerabilities.” 

With the stakes for weather forecasts becoming ever higher, demand for more reliable and tailored insights on a hyper-local level has spawned a new crop of startups—with some claiming a level of precision that’s raising eyebrows.

Through a global infrastructure that includes 17 weather satellites, radar, weather balloons, buoys and ground-based weather sensors, NOAA gathers 6.3 billion observations each day. (The agency even has its own uniformed service, as well as a fleet of planes and ships.) It supplements this data with observations from foreign governments and processes it all using numerical weather prediction models. It’s aim is to provide a comprehensive program of forecasts that have historically guided state and local governments, informing everything from the number of snow-plows on the road to life-saving disaster management. 

Until recently, Hoboken was like almost every other city in America. It relied exclusively on NOAA for its weather and hurricane forecasting. Now, it’s also retained the services of a new “predictive weather intelligence platform,” one that claims to be 60% more accurate than its competitors. Called Tomorrow.io, the company has worked with Delta Air Lines, the National Football League and the U.S. Open tennis tournament. Its forecasts, the company claims, even help Ford’s autonomous vehicles escape bad weather and ride-hailing platform Uber manage its fleet.

Hoboken’s initial one-year contract with Tomorrow.io is worth almost $90,000. To Stratton, the chief resilience officer, the appeal of the company is exactitude: Tomorrow.io contends it can synthesize the data he needs right down to a city-block level. And whereas the National Weather Service (which is part of NOAA) might say there’s going to be heavy rain over a three-hour period in your region, Tomorrow.io claims it can break down forecasts to a matter of minutes.

Read More: Flying Straight Into Hurricanes—How NOAA’s Weather Corps Hunts Storms

Tomorrow.io augments free NOAA data with observations that it acquires from local governments. The company also purchases proprietary datasets from private companies who deploy their own sensors. The data is fed into Tomorrow.io’s cloud-based, machine-learning weather prediction models, which have been trained using historical observations, the company said. By aggregating these observations with predictions from their own forecasting model, as well as those of NOAA and the European Centre for Medium-Range Weather Forecasts, Tomorrow.io claims it’s able to produce accurate, hyper-local forecasts.

“The broad nature of NWS forecasting doesn’t necessarily always look at what is going on right in Hoboken,” Stratton said. “Though we’re only a mile squared in size, having the correct information for that mile is the only thing that matters for us.”

By exploiting modern tools such as artificial intelligence, private companies are jumping ahead of the forecasting capabilities of the NWS, said Cliff Mass, a professor of atmospheric sciences at the University of Washington. That said, he’s a bit skeptical of some of the assertions being made.

“I am extremely doubtful of any claims of minute-by-minute, street-level forecasting,” he said. “The observational and modeling technology today is not capable of providing that.” NOAA declined to comment on Tomorrow.io’s claims. 

But Stratton is a believer. He expects the technology will enable Hoboken to pinpoint when and where extreme weather is going to affect specific parts of the city.

“Being underprepared costs lives, whether that’s because your timing is off by a few hours or you underestimate the volume of rainfall,” he said. From a purely financial standpoint, it’ll also allow the city to deploy resources more efficiently, minimizing operational expenses related to overtime and material use.

Because of vulnerabilities in its sewer system, Hoboken’s streets flood when precipitation intensity is greater than .8” per hour, Stratton said. “Officials have programmed Tomorrow.io to send an internal notification if the precipitation is forecast to exceed [that],” he said. If that prediction comes during high tide, the platform triggers “a full-blown emergency response involving the North Hudson Sewerage Authority, police, fire and the Office of Emergency Management.”

Hoboken is by no means the only U.S. city being wooed by private weather forecasters. In an attempt to get ahead of winter storms, Quincy, Mass., signed a deal with Tomorrow.io in May 2021, as did nearby Newton, Brookline and Dedham. Joe Flanagan, Dedham’s director of public works, said the city “made the money back in three storms.” (The company is also part of a pilot program by New York City’s Department of Sanitation.)

In Brockton, Mass., local Director of Operations Pat Hill said he signed up with Tomorrow.io after a trial with a snowstorm approaching last year. His usual forecaster told him he could expect snow between 1 p.m. and 3 p.m., he said. Tomorrow.io said the storm would end precisely at 1:20 p.m. This information (which turned out to be right) enabled him to release people and equipment two hours early. Overall, the program saved $120,000 in manpower costs over three months, Hill said. 

Companies like The Weather Company, AccuWeather and Metro Weather have occupied the private weather space for years. According to Jonathan Porter, vice president of business services at AccuWeather, cities have “very special risks,” not only to people’s safety, but also to businesses, which have to operate. “And they’re realizing very quickly that without the best information, there’s a greater risk to greater liability,” he said.

To be sure, it’s early days when it comes to trading in human experts for AI. Whereas the NWS is staffed by qualified meteorologists and scientists, with decades of experience in identifying weather patterns, the precision of startups like Tomorrow.io is dependent on software that experts like Cliff Mass of the University of Washington and Gregory Jenkins,  a professor in the Department of Meteorology at Penn State, warn remains unproven. 

Even with the all the data in the world, there are “limits to predictability,” said Jenkins. “And the truth is that if you get down to that granular level, is that because you have weather data from that specific location, or is that because of AI?”

©2022 Bloomberg L.P.