Google’s AI Weather Model Outperforms Traditional Forecasting Methods

A new study showcases significant advances in weather and climate modeling by integrating AI technologies with components of standard, physics-based models. The new model, dubbed “NeuralGCM,” has proven to be more accurate than other purely machine learning-based models for one- to 10-day weather forecasts.

It also outperformed the top extended-range models in use today. According to the findings it also excels in projecting climate conditions over decades.

AI’s Inroads into Weather Prediction

The findings highlight the rapid advancement of AI-based weather and climate forecasting. AI models offer significant computational power and timeliness advantages over traditional computer models.

For example, NeuralGCM is open source and designed to run relatively quickly on a laptop, according to study coauthor Stephen Hoyer of Google Research. In contrast, traditional weather forecasting models require hours to run on the world’s most powerful supercomputers.

These computers have to process tens of thousands of lines of code to describe the physical laws governing the atmosphere and oceans.

How does it work?

Developed by scientists from Google Research, Google DeepMind, MIT, Harvard University, and the European Center for Medium-Range Weather Forecasts, the model uses machine learning and a neural network.

This network loosely models neurons in the brain and trains on decades of past weather data. It also incorporates physics equations describing large-scale weather patterns, combining a global circulation model with AI-driven tasks.

Aaron Hill, an assistant meteorology professor at the University of Oklahoma, described one of the biggest novelties of the model as its integration of large-scale physics with AI components. “Other AI forecasting models made by NVIDIA, Microsoft, and other companies do away with the physics altogether,“.

Between the lines

Hill, who was not involved in the study, explained that AI and machine learning techniques are rapidly being adopted in weather and climate research. But, they have yet to be fully implemented in day-to-day public-facing forecasting by agencies like NOAA or its international counterparts.

“Forecasters haven’t built up a reservoir of trust with these AI-based prediction systems yet. They are just now getting their hands on output and looking at prediction fields on a semi-regular basis, and trust is built up over time with new systems,” Hill said. “Forecasters are really good at what they do in part because they understand the strengths and weaknesses of the current models they use, and when some of them do well while others have biases.”

What does it all mean for the future?

Hoyer emphasized the need for public sector agencies to invest more in AI systems. As we all know these systems are developing quickly and showing promise. Though, as he noted, these systems are not yet ready to replace traditional weather and climate models.

“I think it really shocked a lot of people in the field that you can use AI in the guts of the weather and climate simulation engines for all these downstream applications,” Hoyer said of the new study and other recent work.

The integration of AI into weather and climate modeling represents a significant leap forward. It combines the strengths of traditional models with the speed and adaptability of machine learning.

Originally posted on OpenDataScience.com

Read more data science articles on OpenDataScience.com, including tutorials and guides from beginner to advanced levels! Subscribe to our weekly newsletter here and receive the latest news every Thursday. You can also get data science training on-demand wherever you are with our Ai+ Training platform. Interested in attending an ODSC event? Learn more about our upcoming events here.