Google’s new GraphCast AI model can predict the weather 10 days in advance

Google LLC Today Hinge A new AI model, GraphCast, says it can generate weather forecasts faster and more accurately than traditional algorithms.

The neural network is developed by the company Recently formed Google DeepMind Research Unit. DeepMind has made the source code for GraphCast available at github. According to Google, many weather agencies are already using this model to support their work.

GraphCast is a so-called GNN, a specialized neural network optimized for graph processing. A graph is a data structure that can store multiple snippets of information as well as record how they relate to each other. The ability of graphs to store large, interconnected data sets makes them useful for describing complex phenomena such as weather events.

Google trained GraphCast on nearly four decades of weather observations collected by the European Center for Medium-Range Weather Forecasts, or ECMWF. The training data set included information from satellites, radar systems, and other sources. According to Google, ECMWF is one of the weather organizations that has adopted GraphCast to support its work.

GraphCast can forecast the weather up to ten days in advance. Moreover, it does so with a high degree of detail. Google says GraphCast can predict temperature, humidity levels, wind speed, and other variables at dozens of different elevation levels.

In an internal test, the company compared GraphCast with a weather forecasting algorithm called HRES, which is known for its reliability. GraphCast achieved higher accuracy than HRES across more than 90% of the 1,380 weather variables analyzed during testing. The AI ​​system performed best when the mission scope was limited to the troposphere, the part of the atmosphere directly above the Earth’s surface.

“When we restricted the evaluation to the troposphere, the 6-20 km region of the atmosphere closest to the Earth’s surface where accurate prediction is most important, our model outperformed HRES on 99.7% of test variables for future weather,” Google said. Research scientist on the DeepMind team, Remy Lam, explains in detail in a blog post.

During testing, Google researchers also determined that GraphCast could outperform traditional forecasting algorithms in another area: forecasting severe weather events. Although it was not trained to do so, GraphCast was able to predict tornado movements more accurately than HRES. Furthermore, Google believes that the AI ​​system could be useful for predicting floods.

Traditional weather forecasting algorithms require a significant amount of hardware to run because they rely on complex physics equations to generate forecasts. As a result, such algorithms are often deployed on supercomputers. Even with supercomputer resources, generating a forecast can take several hours in some cases.

Google says GraphCast requires much less infrastructure. According to the company, the AI ​​system can generate 10-day weather forecasts using just one instance of Google Cloud TPU v4. Moreover, it is able to generate those predictions in less than a minute.

“Pioneering the use of artificial intelligence in weather forecasting will benefit billions of people in their daily lives,” Lam wrote. “But our broader research is not just about predicting the weather – it is about understanding the broader patterns of our climate. By developing new tools and accelerating research, we hope that AI can empower the global community to address our biggest environmental challenges.”

Image: Pixabay