Fact Checked by 
Google DeepMind has unveiled its latest weather forecasting AI, GenCast, which it claims delivers predictions both faster and more accurately than traditional physics-based simulations. The details of this breakthrough model have been published in the journal Nature, which shows that the ongoing research project is aiming to improve weather forecasting with artificial intelligence.
GenCast was trained using four decades of historical data from the European Centre for Medium-Range Weather Forecasts’ (ECMWF) ERA5 archive. This dataset includes comprehensive measurements of temperature, wind speed, and atmospheric pressure at various altitudes across the globe.
Today in @Nature, we’re presenting GenCast: our new AI weather model which gives us the probabilities of different weather conditions up to 15 days ahead with state-of-the-art accuracy. ☁️⚡
Here’s how the technology works. 🧵https://t.co/PWCNWbQnlU pic.twitter.com/6DTrmn64Jq
— Google DeepMind (@GoogleDeepMind) December 4, 2024
Google DeepMind claims to predict weather faster
In a blog post, the tech giant states that the new AI model could deliver faster, more accurate forecasts up to 15 days ahead. It also revealed up to a 20% improvement over the ECMWF’s ENS forecast.
In the short term, the model is expected to complement rather than replace traditional forecasting methods. Even in a supportive role, it has the potential to give greater accuracy in predicting extreme weather events such as cold snaps, heat waves, and high winds. This could also help energy companies in forecasting wind farm power generation more effectively.
Ilan Price, a research scientist at Google DeepMind, wrote on X: “Overall, GenCast marks something of an inflection point in the advance of AI for weather prediction, with SOTA raw forecasts now coming from AI.
“I think we can expect them to be increasingly incorporated operationally alongside traditional models (and to continue to improve!)”
4. Overall, GenCast marks something of an inflection point in the advance of AI for weather prediction, with SOTA raw forecasts now coming from AI. I think we can expect them to be increasingly incorporated operationally alongside traditional models (and to continue to improve!)
— Ilan Price (@IlanPrice) December 4, 2024
In the paper, Price and co-author Matthew Wilson note that ENS ensemble forecasts at a resolution of 0.2° or 0.1° “take hours on a supercomputer with tens of thousands of processors.”
They add: “It takes a single Google Cloud TPU v5 just 8 minutes to produce one 15-day forecast in GenCast’s ensemble, and every forecast in the ensemble can be generated simultaneously, in parallel.”
This is the second AI-driven weather model released by Google in recent months. In July, the company introduced NeuralGCM, a hybrid model that integrates AI with traditional physics-based techniques employed in existing forecasting tools. While NeuralGCM achieved performance comparable to conventional methods, it required significantly less computational power.
In the same month, ReadWrite reported that it had taken a big step toward bringing AI in line with human capability to solve complicated mathematics by pairing two new systems, known as AlphaProof and AlphaGeometry 2.
Featured image: Ideogram
