Sign In or Create an Account.

By continuing, you agree to the Terms of Service and acknowledge our Privacy Policy

Climate Tech

Even the Best AI Weather Forecasters Still Need NOAA

While they’re getting more accurate all the time, they still rely on data from traditional models — and possibly always will.

•
A robot forecaster.
Heatmap Illustration/Getty Images

The National Oceanic and Atmospheric Administration has had a bruising few weeks. Deep staffing cuts at the hands of Elon Musk’s efficiency crusaders have led to concerns regarding the potential closure of facilities critical to data-gathering and weather-forecasting operations. Meteorologists have warned that this could put lives at risk, while industries that rely on trustworthy, publicly available weather data — from insurance to fishing, shipping, and agriculture — are bracing for impact. While reliable numbers are difficult to come by, the agency appears to have lost on the order of 7% to 10% of its workforce, or more than 1,000 employees. NOAA’s former deputy director, Andrew Rosenberg, wrote that Musk plans to lay off 50% of the agency, while slashing its budget by 30%.

Will that actually happen? Who the heck knows. But what we can look at are the small cracks that are already emerging, and who could step in to fill that void.

One thing that’s certain is that the National Weather Service, a division of NOAA, announced last week that it is suspending operations at a weather balloon launch site in Alaska, due to staffing shortages. The data gathered at this remote outpost helped inform the agency’s weather forecasts, which are relied upon by hundreds of millions of people, as well as many of the world’s largest companies and public agencies.

Perhaps to Musk’s department, this looks like a prime opportunity for the private sector to step up and demonstrate some nimble data gathering prowess — and indeed a startup that I’ve covered before, WindBorne, has already offered its services. The company, which makes advanced weather balloons, has offered to provide NOAA with data from its own Alaska launches for six months, at no cost. WindBorne is also one of a number of private companies creating AI-based weather models that have outperformed NOAA’s traditional, physics-based models on key metrics such as temperature, wind speed and direction, precipitation, humidity, and pressure.

All this raises the question, though, of what kind of role the private sector could and should play in the weather forecasting space overall. If the architects of Project 2025 have their way, NOAA would be “broken up and downsized,” and its National Weather Service division would “fully commercialize its forecasting operations.” If the Trump administration achieves these goals, “the Weather Service would cease to function in a way that it could meet its mandate to protect American life and property,” Daniel Swain, a climate scientist at University of California Agriculture and Natural Resources, told me.

But given that heavyweights like Google, Huawei, and Nvidia are already in the AI-based weather prediction game, along with startups such as WindBorne and Brightband, which is making weather predictions tailored to the needs of specific industries such as insurance, agriculture, or transportation, it wasn’t clear to me whether, if NOAA were to crumble, the accuracy of weather forecasts necessarily would, too. I thought that perhaps Musk, the White House’s most notorious AI enthusiast, might be thinking the same thing. So I asked around.

“There’s actually a very good argument that I think would be very uncontroversial to expand the role of the private sector, even to offload certain parts of the workflow to the private sector,” Swain told me, with regards to NOAA and its adoption — or lack thereof — of AI-based weather forecasting. But what nobody wanted was to get rid of free, publicly available government forecasts completely.

“I don’t want to have to figure out what company to trust. I just want to be able to go and open the National Weather Service and know what’s going on,” John Dean, the CEO and co-founder of WindBorne, told me.

Julian Green, the CEO and co-founder of Brightband, agreed. “The government doesn’t just forecast the weather, but it gives people alerts. And there’s regulation around whether [it tells you that] you should evacuate, or shut your factory down, or so on.” It’s not hard to imagine the ethical quandaries that could arise from a private company with a profit motive deciding who can access potentially life-saving forecasts, and for how much.

WindBorne’s and Brightband’s AI models, as well as those from tech giants such as Google, are significantly less computationally intensive to operate than those from NOAA or the other leading weather forecasting agency, the European Center for Medium-Range Weather Forecasts. These traditional models rely on supercomputers crunching complicated atmospheric equations based on the laws of physics to make their predictions.

But this doesn’t mean the physics-based models are getting replaced by AI now, or potentially ever. Government data and traditional forecasts still make up the backbone of advanced AIs, which are trained on decades of data largely gathered by NOAA satellites, weather balloons, and radar systems, and then interpreted through the lens of standard physics-based models. After training is complete, the AI models can predict what weather patterns will develop, much like ChatGPT predicts the next word in a sequence, but only after being fed a snapshot of initial weather conditions — also pulled from traditional physics-based models.

Essentially, these AI forecasts are built on the backs of the giants, and while their outcomes are hugely promising, they could not exist without that solid foundation. While one day, it might be possible to operate AI forecasting models without relying on traditional models, Dean and Green told me that physics-based models might always be critical for training the AI. So while their companies’ respective models have yielded impressive results, both Dean and Green nixed the idea that their companies could wholly replace the predictions made by the National Weather Service.

All of this is in flux of course, but as Green put it to me in an email, “a good mechanic doesn't throw away good older tools just because you get new tools.” Plus, as Dean explained, there are still conditions under which physics-based models tend to outperform AI, such as “really small-scale and high-res phenomena — let’s say convective events, let’s say severe thunderstorms in the Plains, or tornado formation.”

Even Project 2025’s authors point out that private industry forecasters rely on publicly available NOAA data, though it doesn’t make any reference to AI models or physics models. The document simply says that the agency “should focus on its data-gathering services” and the “efficient delivery of accurate, timely, and unbiased data to the public and to the private sector.”

There are also questions around whether AI models, trained on data from the past, will be able to predict the types of unusual and extreme weather events that are becoming more and more common in a warming world, Swain told me. “Does it fully capture those?” he asked. “There’s a lot of evidence that the answer is no.”

Lastly, NOAA’s weather model, the Global Forecast System, is simply measuring much more than the AI models do today. “It predicts so many different phenomena, like different types of snow, hail, mixing ratios, turbulence,” Dean said. “We’re building up over time to add more and more variables. But for both WindBorne and other models, it’s not the same currently as what GFS does.”

So while the Heritage Foundation might want to delegate all forecasting responsibilities to private companies, the vision I heard from the startups I talked to looked more like a mutually beneficial arrangement than the full commercialization of weather prediction, or even a clean division of labor. “It’s not privatized weather, it’s a public-private partnership,” Dean said of his ideal future, “where you get freely available forecasts from a public institution like NOAA, but they work with our industry to iterate faster and to drive more innovation.”

What everyone seems to want is simply for the government to forecast better, and today that means moving quickly to build AI-based models. NOAA has taken some steps forward, prototyping some models, bolstering its computing capabilities, and even recently partnering with Brightband to optimize its observational data to train AI models. But it remains behind other agencies in this regard. “The Chinese government and the European Center for Medium Range Weather Forecasts have done a far better job at adopting AI-based weather forecasts than NOAA has,” Dean told me. “So something does need to change at NOAA to get them to move faster.”

Indiscriminately laying off hundreds of the agency’s employees may not be the best place to start. But if there’s anything we know Musk loves, it’s AI and private sector ingenuity. So maybe, just maybe, this administration will be able to forge the kind of partnerships that can supercharge federal forecasting, while keeping NOAA’s weather predictions free and open for all. Or maybe we’ll all just be paying the big bucks to figure out when a hurricane is going to hit.

Yellow

You’re out of free articles.

Subscribe to access Heatmap's exclusive climate and energy polling, award-winning journalism, and subscriber-only events. Save $57 on an annual subscription, just $156 $99/year.
To continue reading
Create a free account or sign in to unlock more free articles.
or
Please enter an email address
By continuing, you agree to the Terms of Service and acknowledge our Privacy Policy
Q&A

How to Channel Voters’ Anxieties About Data Centers

A conversation with the Guardrails Alliance’s Shaunna Thomas.

The Q&A subject.
Heatmap Illustration

This week’s Q&A is with Shaunna Thomas of the Democrat-allied super PAC Guardrails Alliance, which describes itself as “not anti-technology” but supportive of AI “rules, accountability, and a government that actually represents the public.” Part of that AI accountability they’re calling for is on data centers.

After the midterms, Guardrails Alliance sent a memorandum to all House and Senate Democratic offices arguing for AI and data center accountability to be a key part of their message and plans heading into the new Congress. Described as a “playbook for how to understand and channel anxieties about AI and data centers,” the document laid out “key narratives” that they said voters who care about these issues want to see reflected and echoed by their politics. They said the push against data centers could help Democrats build more lasting political power.

Keep reading...Show less
Yellow
Hotspots

Consider the Lobstermen Affected by New York’s Empire Wind

Plus more of the week’s biggest development fights.

The United States.
Heatmap Illustration/Getty Images

1. New York City — We have a new must-watch offshore wind lawsuit with potential implications for the entire sector — and this time, it’s from lobstermen seeking damages.

  • In late October, a family of lobstermen sued Equinor in federal court claiming the company’s compensatory mitigation program is failing to adequately make up for alleged negative impacts to lobster fishing grounds in the New York Bight. According to the lawsuit, the wind project irreparably hurt fishing in previously trafficked waters. The suit also claims that Empire Wind construction destroyed lobster fishing pots and seeks at least $75,000 in damages as well as punitive damages and attorneys’ fees.
  • The lawsuit is the first I have seen involving someone in aquaculture suing for damages over negative impacts of an already-constructed wind project, and it carries outsized implications as the Trump administration remands without vacatur the construction and operations plans for other offshore wind farms. This case is far from being decided, but that doesn’t matter: It is precisely the kind of complaint that Trump officials are likely to consider as they mull revisions to operational plans.
  • An Equinor spokesman declined to comment on the case, telling me they don’t speak on pending litigation.

2. Hinds County, Mississippi — The largest proposed solar farm in Mississippi is moving forward after an appeals court rejected arguments from aggrieved local residents that it had been improperly approved.

Keep reading...Show less
Yellow
Spotlight

Data Center Developers Can’t Pay People Enough to Accept Them

Trump’s “data center dividend” appears to have backfired. Our latest poll results help explain why.

Donald Trump and data center protesters.
Heatmap Illustration/Getty Images

This year’s midterm elections called into question the efficacy of a push to pay local residents directly in exchange for building data centers.

Before the elections, I profiled the campaign for politicians to embrace the data center “dividend,” a policy idea modeled off Alaska’s Permanent Fund, which doles out checks to residents from profits off state oil, gas and mineral royalties. The Bitcoin Policy Institute, a tech policy think tank, was one of the first to trumpet the concept, but it was quickly taken up by President Donald Trump, who started raising it in speeches and on social media. “If companies want to make their Data Centers popular, pay the community, and/or individuals within the community, a ‘dividend,’” Trump posted to Truth Social on October 21.

Keep reading...Show less
Yellow