· Desk editor: Adam Whitmore · · Single source

Google AI hurricane model shows mixed accuracy in new study

University at Albany researchers published an independent evaluation of Google DeepMind's WeatherNext Cyclone model performance across tropical cyclones from June 2022 to May 2026.

Model accuracy compared to official forecasts

The study evaluated intensity forecasts for 385 tropical cyclones worldwide between June 2022 and May 2026. Google's AI model proved more accurate than NOAA's state-of-the-art hurricane model at one to five days out. Its performance matched that of human forecasters after the initial day of predictions. The tool also showed fewer false alarms when predicting rapid intensification over the next 24 hours.

Specific weaknesses in regional predictions

As the forecast range expands, researchers observed that the model often predicts weaker storm intensity than actually occurs. Results differed greatly across regions, achieving peak values in the North Atlantic while reaching their lowest point in the eastern North Pacific. The range of possible outcomes was often too narrow, meaning real storm intensity frequently landed outside the model range.

Study origins from class project

Brian Tang co-led the research which started as a graduate course project last year. Recent graduates Jacob Vile and Nour Elgalad expanded the initial work into a full analysis for peer review. The paper was published in Geophysical Research Letters after independent checks confirmed public confidence needs objective looks at strengths.

Reported by one outlet

Only one outlet has published this. Nothing here has been checked against a second report, so read it as that outlet's account and follow the link below for the original.

Reported by

1 independent outlet. Headline as published. Links open the original report.