New Google AI weather and climate model improves accuracy
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Sarah Grillo/Axios
A new study showcases significant advances in weather and climate modeling, by taking AI technologies and fusing them with some components of standard, physics-based models.
Why it matters: Its creators say 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, along with the top extended-range models in use today.
- It also has proven uniquely skillful in projecting climate conditions over decades, according to the findings, published Monday in the journal Nature.
Zoom in: The findings demonstrate how quickly the field of AI-based weather and climate forecasting are advancing.
- AI models have tremendous compute power and timeliness advantages over traditional computer models.
- The new model, for example, is open source, and is designed to be run relatively quickly on a laptop, study coauthor Stephan Hoyer of Google Research told Axios.
- By contrast, traditional weather forecasting models take hours to run on the world's most powerful supercomputers in order to work through tens of thousands of lines of code describing the physical laws of how the atmosphere and oceans work.
How it works: The new model, from scientists at Google Research, Google DeepMind, MIT, Harvard University and the European Center for Medium-Range weather forecasts, uses machine learning and a neural network.
- This loosely models neurons in the brain, to train off of decades of past weather data.
- It also uses physics equations describing large-scale weather patterns, essentially combining a global circulation model, with its physics-intensive approach, with AI-driven tasks.
Aaron Hill, an assistant meteorology professor at the University of Oklahoma, told Axios that one of the biggest "novelties" of the new model is how it keeps some of the large-scale physics and replaces some parts of the modeling with AI.
- Other AI forecasting models made by NVIDIA, Microsoft and other companies do away with the physics altogether, he noted.
Between the lines: Hill, who wasn't involved in the new study, said AI and machine learning techniques are rapidly being adopted in the weather and climate research communities.
- But they haven't yet made the leap into operations, meaning day-to-day, public-facing forecasting at agencies like NOAA or its international counterparts.
- "One important aspect is 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 they're saying: Hoyer, of Google Research, said public sector agencies have come to see that they need to more fully invest in AI systems, which are developing quickly and showing promise, but not to replace their traditional weather and climate models just yet.
- "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.
