Google Develops AI Model for Predicting Tropical Cyclones

Google LLC has unveiled an artificial intelligence model capable of predicting the trajectory and intensity of tropical cyclones days in advance.

This algorithm, developed through a collaboration between Google Research and DeepMind, is accessible via the newly launched platform called Weather Lab.

Traditionally, scientists have relied on physics-based weather forecasting models to predict cyclones. These models generate forecasts by feeding numerous data points into differential equations. However, Google highlights that this approach has certain limitations that can complicate cyclone prediction.

"In physics-based cyclone forecasting, approximations made to meet operational needs mean a single model often struggles to excel in predicting both a cyclone's path and intensity," explained Google researchers in a blog post. "This is because a cyclone's path is controlled by large-scale atmospheric steering currents, whereas its intensity depends on complex turbulent processes inside and outside its compact core."

According to Google, its new AI model overcomes these limitations. The company claims the algorithm can predict cyclone paths and intensities with "state-of-the-art accuracy." Additionally, its forecasts include other details such as how cyclones form, their size, and shape.

Google trained the model using two datasets. The first dataset outlines the trajectories, intensities, and other key characteristics of nearly 5,000 cyclones over the past 45 years. The second dataset comprises historical weather condition information derived from millions of observations.

In internal testing, Google successfully employed the algorithm to predict the paths of four recent cyclones. For two of these storms, the model generated accurate predictions nearly a week in advance. The researchers noted that it could forecast storms up to 15 days ahead by "generating 50 possible scenarios."

The company has made the AI model available to researchers through the new Weather Lab website. It comes alongside two years of historical forecast data and information from traditional physics-based weather prediction algorithms.

"Weather Lab users can explore and compare forecasts from various AI and physics-based models," the researchers elaborated. "Together, these predictions can help meteorological agencies and emergency services experts better anticipate cyclone paths and intensities."

The release of Google's cyclone prediction algorithm follows just two days after NVIDIA Corporation launched its foundational model named cBottle. According to the chipmaker, this AI can simulate global climate at kilometer resolution. It has already been adopted by several research institutions, including the Max Planck Institute for Meteorology and the Allen Institute for Artificial Intelligence.