Artificial intelligence-based software could calculate weather forecasts faster and at a lower cost than existing methods, say scientists cited by the Wall Street Journal. The use of artificial intelligence to predict the weather has evolved over the past five years from an academic notion to operational testing by weather agencies in the US and Europe, as well as companies that provide such information.
In May, Microsoft launched a forecasting tool called Aurora, which produces five-day global air pollution forecasts and 10-day weather forecasts 5,000 times faster than existing models from the National Oceanic and Atmospheric Administration and the European Center for Medium-Term Weather Forecasts.
Companies and labs across the country, including Villanova University, the University of Oklahoma, and a California startup are preparing new weather AIs.
Faster and more accurate forecasts are becoming increasingly important for companies in all industries. “We absolutely need weather forecasts to be very accurate,” says Remi Lam, a research scientist at Google DeepMind, which introduced an AI-based weather model called GraphCast in November.
For decades, meteorologists have developed weather forecasts using equations such as the relationship between atmospheric pressure and prevailing winds from one region to another, or how quickly temperatures change as cold fronts pass.
Scientists supplement these equations with hour-by-hour measurements of the atmosphere and ocean by weather stations, high-altitude balloons, ocean buoys, and satellites. The data is fed into supercomputers that produce what is called numerical weather prediction.
The problem is that small errors in weather measurement or calculation can lead to larger forecast errors. In addition, running complicated Earth weather simulations requires a lot of expensive computing time.
AI algorithms look for patterns in weather data rather than solving equations like supercomputers do. Pattern-finding algorithms are trained on decades of weather data to predict what will happen in the coming days.
“All these AI tools do is recognize patterns,” says Paris Perdikaris, principal investigator of the Aurora project at Microsoft Research. “And they’re very good at it.”
According to Perdikaris, the researchers trained Aurora with a huge amount of historical weather data to make these predictions, about 16 times more data than was used to train the latest version of the AI-powered ChatGPT chatbot.
Microsoft expects to make Aurora available to the public in the months following testing.
