AI system outperforms humans in global weather forecasting
In a breakthrough for artificial intelligenceresearchers at Google deep mind have developed an AI system called GraphCast that can predict global weather up to 10 days into the future with more accuracy than traditional forecasting methods. The results were published this week in the journal Science.
According to a recent announcement, GraphCast was more accurate than the current leading weather forecast system run by the European Center for Medium-Range Weather Forecasts (ECMWF) – in more than 90% of the 1,380 evaluation metrics tested. These measurements included temperature, pressure, wind speed and direction, and humidity at different atmospheric levels.
GraphCast works using a machine learning technique called graph neural networks.
It was trained on more than 40 years of past ECMWF weather data to learn how weather systems develop and move around the world. Once trained, GraphCast only needs the current state of the atmosphere and the state six hours ago as inputs to generate a 10-day global forecast in about a minute on a single cloud computer.
This method is much faster, cheaper and more energy efficient than the traditional numerical weather prediction approach used by national forecast centers like ECMWF. This technique relies on solving complex physical equations on supercomputers, which takes hours of computational time and energy.
Matthew Chantry, an expert at ECMWF, confirmed that GraphCast consistently outperformed other AI weather models from companies like Huawei and Nvidia. He believes this marks an important turning point for AI in meteorology, with systems advancing “much sooner and more impressively than expected”.
DeepMind researchers point out that GraphCast accurately predicted the arrival of Hurricane Lee in Nova Scotia nine days in advance, compared to just six days for conventional methods. This gave people three extra days to prepare.
GraphCast failed to outperform traditional models in predicting the rapid intensification of Hurricane Otis off the Pacific coast of Mexico.
While promising, experts note that AI models like GraphCast might struggle to account for climate change since they are trained on historical data. ECMWF plans to develop a hybrid approach, combining AI forecasts with physical weather models. The UK Met Office recently announced similar plans, believing that this combined technique would provide the most robust forecasts in the era of climate change.
Featured Image Credit: Photo by Josh Hild; Pixels; THANKS!
