Google Releases Open-Source Weather Model to Improve Hurricane Predictions

Google has announced its WeatherNext AI model to improve hurricane warnings, which it claims can generate a single 15-day forecast in under a minute.

The company developed this AI after recognizing that tropical cyclones, also known as hurricanes or typhoons, are among the most destructive weather phenomena on Earth, responsible for more than 700,000 deaths and $1.4 trillion in economic losses globally over the past 50 years.

According to Google, these types of weather events are uniquely difficult to predict due to global atmospheric currents being traditionally analyzed using coarser global models. To address this problem, WeatherNext trained on nearly 20 terabytes worth of global atmospheric data and used historic information collected by the International Best Track Archive for Climate Stewardship.

The company boasts that its tech has had real-world impact, having been used to help the NHC forecast Hurricane Melissa, allowing the agency to issue advanced warnings and enable teams more time to prepare.

Now, Google is making WeatherNext AI open-source, allowing developers to study the code and build something of their own based off of it. The company hopes this will accelerate progress across the global weather community and empower meteorological agencies, researchers, and nonprofits to better predict weather events and make key decisions to protect lives and infrastructure.