When Helene hit Florida earlier this year, it was the worst hurricane to hit the US mainland since Katrina in 2005, killing 234 people. It is natural disasters like this and their increasing intensity due to climate change that have prompted scientists to develop. more accurate weather forecasting systems. On Wednesday, Google’s DeepMind division announced what may go down as the most significant advance in the field in its nearly eight decades of operation.
in an article about Google Keyword BlogDeepMind’s Ilan Price and Matthew Wilson detailed the company’s latest artificial intelligence agent, GenCast. According to DeepMind, GenCast is not only better at predicting daily and extreme weather than previous AI weather software, but also outperforms the best forecasting system currently in use. ECMWF). In tests comparing the two systems’ 15-day forecasts for weather in 2019, GenCast was more accurate than ECMWF’s ENS system 97.2 percent of the time, on average. With production times of more than 36 hours, DeepMind’s was even better at 99.8 percent more accurate.
“I’m a little reluctant to say it, but it feels like we’ve made decades worth of improvements in one year,” said Rémi Lam, lead scientist on DeepMind’s previous AI weather program. he said The New York Times. “We’re seeing really, really fast progress.”
GenCast is a diffusion model, the same technology that powers Google’s generative AI tools. DeepMind trained the program on nearly 40 years of high-quality weather data curated by the European Center for Medium-Range Weather Forecasts. The predictions produced by the new model are probabilistic, meaning they consider a range of possibilities, which are then expressed as percentages. Probabilistic models are considered more nuanced and useful than their deterministic counterparts, offering the best estimates of what the weather might be like on a given day. The first one is more difficult to create and calculate.
Indeed, what is perhaps most striking about GenCast is that it requires significantly less computing power than traditional physics-based ensemble forecasts such as ENS. According to Google, its one and only TPU v5 tensor processing facilities can generate a 15-day GenCast forecast in eight minutes. In contrast, it may take a supercomputer with tens of thousands of processors to produce a physics-based prediction.
Of course, GenCast isn’t perfect. One area where software can make better predictions is hurricane intensity, though the DeepMind team The Times he was confident that he could find solutions to the agent’s current shortcomings. In the meantime, Google is making GenCast open source with sample code for the tool Available on GitHub. GenCast predictions are also coming to Google Earth soon.
