How AI is making weather forecasts more accurate by Cynthia McCormick Hibbert May 18, 2023 Share Mastodon Facebook LinkedIn Twitter Kate Duffy and Thomas Vandal work on their weather prediction startup, Zeus AI on May 16, 2023. Photo by Matthew Modoono/Northeastern University When people aren’t complaining about the weather, they are deploring inaccurate weather forecasts that lead to rain on a planned beach day or a windstorm on a still day. Thomas Vandal and Kate Duffy worked as NASA scientists after earning doctoral degrees in interdisciplinary engineering from Northeastern. They intend to improve on short-term forecasting with a new startup that uses AI and machine learning to predict weather patterns. The startup, Zeus AI, draws on the enormous amount of data provided by the latest generation of government satellites—atmospheric winds, water vapors, temperature changes and cloud cover that influence weather across the globe. “It’s a huge volume of data,” says Duffy, who is the chief product officer for Zeus. “A lot of it isn’t used in current weather prediction models.” “We’re looking at retrieving more dense and complete information about the state of the atmosphere from geostationary data that can then be incorporated into weather models,” she says. The latest generation of NASA and NOAA geostationary satellites are known as the GOES-R series satellites. They are “substantially better than the previous generation,” says Vandal, who is the CEO of Zeus. He says, “this enables way more applications” using machine learning and AI to crunch the data, see patterns and make predictions. Kate Duffy and Thomas Vandal work on their weather prediction startup, Zeus AI on May 16, 2023. Photo by Matthew Modoono/Northeastern University Kate Duffy and Thomas Vandal work on their weather prediction startup, Zeus AI on May 16, 2023. Photo by Matthew Modoono/Northeastern University Photos by Matthew Modoono/Northeastern University “Traditional weather forecasting systems are insanely expensive to run—they run on the biggest supercomputers in the world, basically,” Vandal says. “And they’re actually not able to take in this high density data. “This is where machine learning comes in. We’re able to do this much less expensively than the government systems because we can do it on a single machine. We’re able to learn quickly from the data that already exists,” Vandal says. The roots of Zeus AI can be traced back to Northeastern’s Sustainability and Data Sciences Lab, director Auroop Ganguly says. Ganguly says the two, who recently obtained Small Business Innovation Research phase II funding from NASA to launch their startup, showed early on an extraordinary ability to do interdisciplinary research combining both AI and climate science. “They have won a best-paper award at a highly-selective data science conference, published in top climate and machine learning journals and conferences, as well as in high impact interdisciplinary journals such as Nature Climate Change and Nature Communications,” Ganguly says. “Their work has been highlighted as path breaking in comments and review articles in Nature.” Ganguly says Vandal “is among the top few who understands machine learning and AI for weather and climate analytics,” adding that Duffy understands “critical gaps” in AI in a way that benefits the earth and environmental sciences and engineering. .gifs courtesy Zeus AI “Together they make a fascinating team,” Ganguly says, who is also co-director of Northeastern’s Global Resilience Institute and director of AI for climate and sustainability within the Institute of Experiential AI. Vandal and Duffy estimate that the Zeus AI model can generate more accurate forecasts faster than U.S. government models such as the High-Resolution Rapid Refresh by the National Oceanic and Atmospheric Administration. But don’t expect Zeus to be used on favorite news weather stations any time soon. The intended clients of the new startup include energy markets and energy traders, and the daily pricing of energy based on demands for things such as air conditioning and heating materials, Vandal says. Renewable energy companies, such as in wind and solar, can benefit. More accurate weather models can also help with green energy efforts by predicting how weather conditions are impacting solar and wind production, so there is less reliance on thermal energy storage as a backup, Duffy says. “A better forecast could help more efficiently integrate renewables into the energy system and lower the cost,” she says. Vandal says Zeus AI promises a unique way of efficiently processing satellite data using machine learning. “NASA and NOAA provide all this data for free to the public,” he says, adding that it takes a system like Zeus, named after the Greek god of weather, to use it to make accurate predictions. Ganguly says Vandal and Duffy are among a growing number of former Sustainability and Data Sciences Lab students launching entrepreneurial activities in the field of climate change and resilience. “Ph.D. alumnus Evan Kodra led a successful climate analytics startup for cities and urban bond markets, which was reported by Northeastern Global News and the Wall Street Journal, among others,” Ganguly says. The startup, originally called risQ, was acquired by Fortune 500 company Intercontinental Exchange over a year ago and was launched from the Sustainability and Data Sciences Lab with Small Business Innovation Research funding from the National Science Foundation. Ganguly called Vandal, Duffy and Kodra an example of the “best brains of the younger generation (leveraging) solutions such as data analytics and AI to address what have been called the defining grand challenges of our age.” Cynthia McCormick Hibbert is a Northeastern Global News reporter. Email her at firstname.lastname@example.org or contact her on Twitter @HibbertCynthia.