How AI struggles to forecast hurricane intensity spikes

KUALA LUMPUR, 27 SEPTEMBER 2026 —

Artificial intelligence models that predict weather have reached a level of accuracy that rivals the best physics‑based systems, yet they still struggle to forecast the rapid intensification of hurricanes.

The study, which was republished from The Conversation under a Creative Commons licence, explains that AI has benefited from three main drivers: the accumulation of vast weather datasets, advances in machine‑learning architectures, and unprecedented computing power. These factors have enabled global AI weather models to produce forecasts that match the performance of leading physics‑based prediction systems.

However, the same progress has not translated to the regional scale, where forecasting extreme events such as hurricanes remains a challenge.

Hurricanes often undergo rapid intensification, strengthening from a weak tropical storm into a destructive Category 5 system within hours. The article cites Hurricane Polo, which off the Pacific coast of Mexico intensified from a tropical storm on 21 September 2026 to a Category 5 hurricane in 24 hours, with winds reaching 180 mph (290 km h⁻¹).

It also references Hurricane Michael, which in 2018 grew into a Category 5 storm just before striking Tyndall Air Force Base and Mexico Beach, Florida, leaving communities with insufficient time to evacuate. These examples illustrate why accurate intensity forecasts are critical for public safety.

Unlike global weather forecasts, which rely on broad‑scale atmospheric patterns, hurricane intensity forecasts are considered a regional problem. They focus on extreme events that develop rapidly or move quickly over short periods. Capturing such behaviour in AI models requires data with far greater detail than the current global datasets provide.

When scientists train AI models to predict hurricane intensity, they typically use two sources of data. The first is direct observations, including rainfall, near‑surface temperature, wind speed and other variables collected from weather stations, radars, buoys and satellites. These observations can be detailed but are unevenly distributed, especially over the open ocean where most of the critical stages of hurricane development occur.

Modern satellites help fill some gaps, but they can only estimate parts of the rainfall, surface winds or cloud‑top temperatures because of limits in satellite coverage. Even the best observational systems provide only a partial view of hurricanes at any point in time.

The second source of training data comes from weather model simulations, which combine atmospheric conditions and physical knowledge to provide a three‑dimensional picture of the atmosphere at high resolution. However, these simulated data are not perfect because all computer models contain approximations and uncertainties arising from incomplete knowledge of Earth’s atmosphere. Fine‑scale processes that influence hurricane intensity are often beyond the resolution of these models.

The article argues that even if future technology could measure every part of thousands of storms around the world every second, AI would still face fundamental limits due to chaos. Tiny differences in the initial state of a hurricane can grow rapidly over time. The study’s authors suggest that hurricanes may contain an element of chaos that prevents AI models from accurately predicting intensity at long forecast times.

Hurricanes have a potential intensity, the maximum strength they can reach in a given environment. Warm ocean water fuels a hurricane’s intensity, while wind shear can slow its development. If the ocean temperature rises, the potential intensity increases. Small disturbances also cause intensity to fluctuate, and the warmer the ocean surface, the greater the fluctuations.

Recent studies propose that these fluctuations are not purely random but occur within a chaotic attractor—a set of possible storm states within which the hurricane can evolve unpredictably. Although the existence of such a chaotic intensity attractor has not yet been fully established, it presents a fundamental dilemma for training AI models to predict hurricane intensity.

On one hand, scientists want AI models to make the most accurate predictions possible. During training, the goal is to minimize the difference between the forecast and what actually happens until the AI model achieves the smallest possible error. On the other hand, the AI model must also capture the hurricane’s intrinsic chaos. If an AI model can capture this chaos, its error cannot be reduced indefinitely.

An AI model trained to minimize forecast error may therefore learn the most likely evolution of a hurricane while smoothing out unpredictable fluctuations. In this regard, the two goals compete with one another.

Because data always contain some uncertainty, the rules an AI model learns are only approximations. The accuracy of hurricane intensity forecasts will therefore get worse after just a few days. The challenge for AI models predicting hurricane intensity is not just about obtaining more data, building better neural networks or deploying faster computers. It is also about understanding hurricane behaviour and how chaos in intensity emerges.

Both factors dictate whether AI models can learn what is predictable and what is unpredictable. That distinction not only puts a cap on the accuracy of current hurricane intensity forecasts but also determines the next generation of weather forecasting and evaluation systems. Future systems should focus on a range of possible hurricane intensities and their probabilities instead of a single intensity number.

The article concludes that AI’s struggle to predict hurricane intensity is rooted in sparse ocean observations, imperfect simulation data and the chaotic nature of tropical cyclones. Addressing these issues will require not only better data and computing power but also a deeper understanding of the physics that govern hurricane development.