Exclusive: AI weather firm Atmo teams up with Tuvalu on climate resilience

Zoom in: Atmo is a San Francisco-based AI weather forecasting firm whose models are trained on historical weather conditions. The models can be run far faster and on cheaper machines than the supercomputers used by the National Weather Service and other agencies that predict the weather.

• Right now, there's a dearth of weather observations and relatively poor performance by U.S., European and other global weather models at the local level in the South Pacific.
• It means that the residents of Tuvalu are often caught off guard by storms, Atmo CEO Alex Levy tells Axios in an interview.
• Given how low-lying the country is — the average height is less than 10 feet above sea level — these storms can cause extensive flooding.

Inside the room: Levy first met Tuvalu ministers and learned about their challenges with sea level rise and extreme weather while at the COP27 climate summit in Sharm El-Sheikh, Egypt.

State of play: Tuvalu officials and Levy are set to sign an agreement Thursday afternoon in New York to formalize a partnership in which Atmo's high-resolution AI-based forecast model will provide zero to 10-day forecasts for the island nation.

Between the lines: The partnership is unique in that it brings together a next-generation private sector weather company and one of the most vulnerable countries on the planet to address the ravages of climate change.

• It also fits into the framework of what U.N. Secretary-General António Guterres has been seeking: the spreading of early warning technologies to countries that have traditionally lacked such infrastructure.
• However, Levy clarified this partnership is not formally the result of the UN's "Early Warnings for All" initiative.

How it works: Levy says historically, meteorology has developed based on a "best and the rest" framework. Wealthy countries built and deployed expensive global forecasting models, such as the National Weather Service's Global Forecast System.

• Poorer nations lacked access to their own models that might be better suited to their needs.
• "There was no financial structure that would allow even a middle-income country to design and operate a finely tuned forecast for itself under the previous way of doing it," Levy told Axios.

The intrigue: AI meteorology is fundamentally changing the game, he said, since forecasting models can now be highly focused on particular needs or meteorological hazards.

• Also, AI models such as Atmos' can run at about 40,000 times faster than the global models agencies like the National Weather Service still rely on, and they can refresh in 8 seconds rather than more than four hours, he said.
• "So that then allows you to create, country by country, highly tuned and trained forecasts that are calibrated to that country's historical data," Levy said.

Yes, but: There are fundamental questions facing AI forecasting companies such as Atmo, mainly revolving around their ability to foresee unprecedented events, since their models are trained on historical observations and outcomes.

• Levy said that concern is understandable but not proving out in the model's forecasts, which have been catching such occurrences in tests.
• Machine-learning techniques, Levy said, are picking up on how the atmosphere works and becoming less likely to miss a major outlier event.

The intrigue: The partnership between Tuvalu and Atmo was nurtured by a U.N.-hosted entity, the Global Centre for Climate Mobility, which works to foster climate adaptation for countries most at risk from climate change.

What they're saying: "Innovation and technology to address the adverse impacts of climate change should not be limited to the richest," Kamal Amakrane, managing director of the global center and the climate envoy of the president of the UN General Assembly, tells Axios in an interview.

• It "should not be limited to the most powerful, and should not only be limited to those who can afford it."