Google DeepMind has introduced a new AI model for weather forecasting called GenCast, which is already demonstrating high accuracy. According to a study published in Nature, GenCast outperformed one of the leading forecasting models, ENS, used by the European Centre for Medium-Range Weather Forecasts (ECMWF), in 97.2% of cases.
GenCast uses machine learning to analyze weather data from 1979 to 2018 in order to predict future weather conditions. This differs from traditional models, which rely on supercomputers to solve complex equations of atmospheric physics. “Weather affects every aspect of our lives… it’s one of the great scientific challenges,” notes DeepMind senior scientist Ilan Price.
Speed is one of GenCast’s advantages: it can generate a 15-day forecast in just eight minutes using a single Google Cloud TPU v5. This is much faster than physical models, which can take several hours. “Computationally, it’s orders of magnitude more expensive to run traditional forecasts compared to a model like GenCast,” Price notes.
Despite its successes, GenCast still has room for improvement, particularly in increasing its resolution. It’s important to note that GenCast has so far been tested on an older version of ENS. However, DeepMind conducted similar studies on data from 2020 to 2022 and obtained comparable results.
DeepMind has released the GenCast model code as open source, allowing experts to independently test its capabilities. Price believes that this model and similar advanced AI models can be used alongside traditional models in the real world, which will help build trust and confidence in their use.
Google’s innovative developments in artificial intelligence continue to impress. In particular, the company recently opened access to the Veo video model for business, demonstrating the growing potential of AI technologies across various industries. Meanwhile, Google is already preparing to release its new Gemini 2.0 model in December, which promises to further enhance AI capabilities.
