Google is working with the National Hurricane Center this cyclone season to improve storm forecasting, and it’s offering more than just support. You see, the company is rolling out a new AI model built to predict the path, size, and intensity of cyclones with impressive speed and accuracy.
The model comes from Google DeepMind and Google Research. Instead of relying on the usual physics-based simulations that run on supercomputers, this one uses machine learning. It has been trained on both general weather patterns and cyclone-specific data, allowing it to make near-instant predictions. That kind of speed matters when lives are on the line.
Google has tested weather models in the past like GenCast and GraphCast. They showed promise but weren’t built specifically for cyclones. Their lower resolution and weak intensity predictions meant they weren’t fully trusted by meteorologists. This time, the focus is entirely on tropical storms, and the difference shows.
The new model produces 50 different possible outcomes for each storm, giving forecasters a much broader sense of what could happen. In testing, it predicted several storms — including Cyclones Honde, Garance, Jude, and Ivone — nearly a week before they officially formed. That kind of early warning could mean the difference between chaos and preparation.
Rather than simply handing off the tool, Google is actively working with the National Hurricane Center to refine the model. One result of that collaboration is a feature called expert mode. It shows where storms might form by highlighting clusters of small circles on a global map. Each circle represents a small chance of development, letting forecasters explore potential outcomes early in the storm cycle.
Cyclones are notoriously hard to predict. Their wind speeds and chaotic nature make traditional forecasting models struggle. To address that, the team created a probabilistic model that works in a single step, introducing random variations to generate multiple outcomes at once. That gives forecasters more information, faster.
In one real-world example, the model correctly anticipated that Cyclone Alfred would weaken to a tropical storm and make landfall near Brisbane, Australia, seven days before it happened. That’s not just accurate. It’s useful.
Google is also launching something called Weather Lab, a platform where people can view both real-time and historical forecasts. Trusted testers have been using it for the past couple of months, offering feedback on everything from interface tweaks to the kind of data that matters most during a storm.
Personally, I think this is the kind of AI we should be talking about more often. I love when tech helps people stay safe, rather than just pushing another product or service. When artificial intelligence is used to save lives, it proves it can actually be meaningful.
