🌪️ 3D ‘X-ray’ of a hurricane with AI

Hurricane Melissa struck Jamaica and the Caribbean on Tuesday with tragic intensity, leaving thousands displaced and over 50 lives lost. With gusts over 250 kph and a 3m storm surge, Melissa was one of the three most intense storms to ever make landfall. 

At the FDL Earth Systems Lab (ESLab.ai) in partnership with ESA, Google NVIDIA and SCAN computers, we’ve been researching how we can use #AI to better understand the structure of tropical cyclones, particularly at the key ‘intensification moment’.

Why does this help forecasters? If a storm suddenly intensifies, it can catch thousands unprepared. But false positives, when warnings don’t come to pass, can create complacency.

Accurately forecasting how a storm might behave at landfall is a key challenge in weather forecasting..

This 3D tomographic nowcast of hurricane Melissa on 27 October 2025 highlights the storm’s inner dynamics eight hours before it made landfall in Jamaica. The eye is perfectly defined, the eyewall nearly circular, which are textbook signs of a Category 5 hurricane at peak intensity. The volumetric model reveals the intense water content (measured in density, droplet size and ice water content) sweeping towards Jamaica on the east side of the eye - carrying a year’s worth of rain (about 750mm). 

🛰️ Why 3D tomography matters…

Conventional weather satellites, like NOAA’s Geostationary Operational Environmental Satellites (GOES) image clouds and tropical cyclones at high temporal cadence from the top, but can’t directly probe their 3D structure. Polar-orbiting radar satellites (like NASA’s CloudSat or ESA’s EarthCARE) can penetrate into clouds to measure vertical structure, but observations are spatially and temporally sparse. Over the course of CloudSat’s 20-year mission, it only passed over ~100 hurricanes within 250 km of the center, and many intense hurricanes were missed. Direct observations of vertical structure, especially of key microphysical parameters like ice and water content, could have helped forecasters better predict if and when hurricanes intensify, and which track they might follow.

Through the use of AI, we can explore Melissa volumetrically and create “virtual” satellite observations; quantifying visual features such as eye diameter, symmetry, eyewall radar reflectivity, and ice water content to give a detailed view of the inner workings of the hurricane for the first time.  

⚙️ New capabilities unlocked: 

Continuous monitoring of intense storms in 3D can:

• Harness the combined power of different satellites to fill in measurement gaps

• Point to rapid intensification hours sooner.

• Provide interpretable visuals for forecasters and emergency managers by showing where the mass of rain and ice is stored within the cyclone. 

• Power real-time dashboards that flag when storms are at their most extreme. 

• Provide insights for better understanding the physical processes that lead to rapid intensification.

What’s more, our model can generate 3D reconstructions in seconds and allow experts to explore and “fly through” the storm. 

🌍 The big picture

As hurricanes and cyclones become stronger and more erratic, AI nowcasting like this offers a new layer of situational awareness, pushing toward faster, clearer, and more reliable alerts for those in harm’s way.

Many thanks to the amazing ESL “3D Clouds for Extremes” team for this incredible analysis. 

You can read more about this AI research product and some of the other amazing outcomes of FDL Earth System Lab here: AI4EO Research Outcomes 2025