AI Predicts Path of Hurricane Helene Better Than Weather Models
November 17, 2026
Hurricane Helene made landfall as a Category 4 hurricane in September 2024, causing tens of billions of dollars in damage and killing over 250 people, with North Carolina particularly devastated.
Research funded by the National Science Foundation compared conventional weather models with artificial intelligence to analyze how each would predict the path of the storm. The scientists found that by analyzing weather data, some AI models could accurately re-create Helene’s path up to a week in advance and came within 85 miles of the actual track — a level of accuracy that, in several cases, outperformed traditional physics-based models.
What makes this research especially noteworthy is that these AI systems don’t rely on solving the complex atmospheric physics equations used by conventional forecast models. Instead, they use neural-network approaches similar to large language models (LLMs), “learning” from decades of global weather data to recognize patterns and generate fast, highly accurate predictions.
Four different AI models produced more accurate track and intensity forecasts further in advance than traditional physics-based tools.
The researchers also found that the “brown ocean effect” was a critical factor in making Helene’s rainfall so catastrophic. This occurs when warm, saturated soils act as a substitute ocean for hurricanes, providing the moisture and energy needed to fuel intense rainfall for 2-3 days after landfall — even hundreds of miles inland.
Research by Hassan Dashtian, research assistant professor; Dev Niyogi, professor; Manmeet Singh, postdoctoral fellow; Alka Tiwari, postdoctoral fellow; Sasanka Talukdar, graduate student
Bureau of Economic Geology; Department of Earth and Planetary Sciences