Google GraphCast AI predicts weather for next 10 days in under a minute

GraphCast became a weather genius by learning its weather tricks using a "graph neural network" and studying over 40 years of historical weather data from ECMWF. When given the current weather and some hints from six hours ago, GraphCast can create a 10-day weather prediction faster than a person can say ‘Google.’

The AI model will also give information about cyclones, floods, and extreme temperatures. This means it can give us a heads-up before these events happen, helping us be better prepared and potentially saving lives.

The team has open-sourced the model, making its code available to scientists and forecasters worldwide.

Weather prediction is a complex task crucial for various decision-making processes. Traditional methods, like Numerical Weather Prediction (NWP), use physics equations translated into computer algorithms running on supercomputers. While effective, it demands expertise and significant resources. 

Enter GraphCast.

But it has its limitations

The team wrote in their study that their method doesn’t deal with uncertainty in weather predictions. Their research mainly focused on specific forecasts and compared them to HRES. As we know, things can get uncertain in real weather scenarios, especially for extreme events, and that aspect has not been taken into consideration.

GraphCast joins Google’s weather prediction family. Its advanced systems like Nowcasting predict weather up to 90 minutes ahead, while MetNet-3 makes accurate 24-hour forecasts in the US and Europe.

Using AI in weather forecasting isn't just about knowing if it'll rain tomorrow; it's a big deal for everyone. Google is leading the way in helping billions of people by using AI to understand not only daily weather but also bigger climate patterns. 

“By developing new tools and accelerating research, we hope AI can empower the global community to tackle our greatest environmental challenges,” wrote Remi Lam, one of the authors of the study, in a blog post.

Study abstract:

Global medium-range weather forecasting is critical to decision-making across many social and economic domains. Traditional numerical weather prediction uses increased compute resources to improve forecast accuracy, but does not directly use historical weather data to improve the underlying model. Here, we introduce “GraphCast,” a machine learning-based method trained directly from reanalysis data. It predicts hundreds of weather variables, over 10 days at 0.25° resolution globally, in under one minute. GraphCast significantly outperforms the most accurate operational deterministic systems on 90% of 1380 verification targets, and its forecasts support better severe event prediction, including tropical cyclones tracking, atmospheric rivers, and extreme temperatures. GraphCast is a key advance in accurate and efficient weather forecasting, and helps realize the promise of machine learning for modeling complex dynamical systems.