Tropical cyclones known as hurricanes or typhoons in different parts of the world are among the most dangerous natural disasters on Earth. In the past 50 years alone, these storms have caused a staggering $1.4 trillion in damages worldwide, devastating communities and taking countless lives. But now, thanks to a new breakthrough by Google DeepMind and Google Research, predicting these deadly storms could become faster and more accurate.
On Thursday, Google launched Weather Lab, an interactive online platform that showcases the latest experimental weather prediction models powered by Artificial Intelligence (AI). Among these tools is a brand-new AI-based tropical cyclone model that can forecast a cyclone’s formation, track, strength, size, and shape up to 15 days in advance.
What makes this model special is its ability to produce 50 possible cyclone scenarios rather than a single outcome. This gives weather experts a fuller picture of potential threats and allows communities to prepare better potentially saving lives and reducing damages.
Better Than Traditional Models?
In early tests, Google’s AI model has matched and sometimes outperformed traditional physics-based forecasting methods, especially for predicting the cyclone’s path and intensity. This is important because while current models are good at tracking where storms may go, they often struggle with predicting how strong they will become.
To ensure accuracy, Google is working closely with top weather organizations, including the U.S. National Hurricane Center (NHC). The NHC is now testing these AI predictions alongside its usual tools to assess their usefulness during real storms in the Atlantic and East Pacific regions.
A Single Model Solving a Big Problem
Traditionally, forecasters use two types of models: one to predict a cyclone’s track across vast ocean currents, and another high-resolution model to guess its strength and intensity. But these models can’t do both well at the same time.
Google’s AI model changes that. By using a massive database of 45 years of past cyclone data, plus global weather patterns, the system can predict both track and intensity in a single process and with impressive results.
In fact, when tested on storms from 2023 and 2024, the AI model could predict a cyclone’s location five days in advance with an average error of 140 kilometers less than the best European weather model (ECMWF’s ENS). This is like getting tomorrow’s accuracy but five days ahead a jump that usually takes scientists more than a decade to achieve.
The model is also performing better at estimating how strong storms will get a major challenge for forecasters beating the NOAA’s advanced HAFS system in initial tests.
Real-Time Tools for Researchers and Experts
Through Weather Lab, Google is offering both live and historical data on tropical cyclones. Weather scientists and agencies worldwide including the UK Met Office, University of Tokyo, and Japan’s Weathernews Inc. are already using the platform to test the AI model’s performance.
However, Google warns that this is still an experimental tool. It is meant for researchers and not for the public to rely on for real-time safety decisions. Official cyclone warnings will still come from national weather services.
A Step Towards Safer Communities
Experts are hopeful that AI-powered forecasting could become an important tool for reducing the impact of deadly storms. Dr. Kate Musgrave, a scientist at Colorado State University’s CIRA research center, said the AI model has shown “comparable or greater skill than the best current models” in predicting cyclone paths and strength. She added that real-time testing during the 2025 hurricane season will give a clearer picture of how well it works in practice.
Google’s Weather Lab is part of its larger WeatherNext research project, aimed at bringing cutting-edge AI to climate forecasting. As extreme weather events become more frequent because of climate change, better prediction tools like this could make all the difference in preparing communities for what lies ahead.