Google DeepMind has developed a new artificial intelligence (AI) weather forecasting model, GenCast, that outperforms traditional meteorological methods in predicting weather out to 15 days and is also more accurate at predicting extreme weather events.
GenCast’s AI model considers the likelihood of multiple scenarios to accurately estimate trends, from wind power generation to the movement of tropical cyclones. GenCast’s probabilistic method is a new frontier in the use of AI to provide better, faster daily weather forecasts. This approach is increasingly being used by major weather services, writes the Financial Times.
«This marks something of an inflection point in the development of AI for weather forecasting, as modern raw forecasts now come from machine learning models,” said Ilan Price, research scientist at Google DeepMind. He added that GenCast could be incorporated into operational weather forecasting systems, allowing meteorologists to better understand trends and prepare for upcoming weather events.
What’s new about GenCast’s approach compared to previous machine learning models is its use of so-called ensemble forecasts representing different outcomes, a technique used in modern traditional weather forecasting. GenCast was trained using the European Center for Medium-Range Weather Forecasting (ECMWF) database, which has been accumulated over four decades.
According to a publication in Nature, the GenCast model outperformed the ECMWF’s 15-day forecast on 97.2% of 1,320 variables such as temperature, wind speed and humidity. Thus, it surpassed the AI model GraphCast from Google DeepMind, introduced last year, in accuracy and coverage. GraphCast outperformed ECMWF forecasts 3 to 10 days ahead in about 90% of the metrics.
AI weather forecasting models are much faster than standard forecasting methods, which rely on enormous computing power to process the data. GenCast can generate its forecast in as little as eight minutes, whereas traditional methods take hours to produce a forecast.
According to the researchers, GenCast’s AI model can be further improved in its ability to predict the intensity of major storms. The resolution of its data may also be increased to match the updates made this year by ECMWF.
ECMWF called the development of GenCast “an important milestone in the development of weather forecasting.” The center also said it has integrated “key components” of the GenCast approach into a version of its AI ensemble forecasting system available since June.