The artificial intelligence (AI) tool predicted that Hurricane Lee would land in Nova Scotia nine days before it happened – three days earlier than traditional methods.
The weather forecasting AI algorithm developed by Google’s DeepMind, known as GraphCast, has outperformed the European Centre for Medium-Range Weather Forecasts (ECMWF) model on more than 90 per cent of 1,380 metrics, including temperature, pressure, wind speed and direction, and humidity at different levels of the atmosphere.
The machine-learning model takes less than a minute to make 10-day weather forecasts worldwide on a desktop computer, and is more precise than other approaches.
“GraphCast currently is leading the race among the AI models,” said computer scientist Aditya Grover at the University of California, Los Angeles.
Traditional weather forecasts use numerical weather prediction (NWP), which uses mathematical models based on physical principles. These tools rely on data from thousands of weather stations at different levels of the atmosphere around the globe. For this reason, they require vast amounts of computing power.
In contrast, GraphCast uses machine learning to digest vast quantities of historical data and predict how weather patterns will evolve. The AI can reportedly produce more accurate forecasts than traditional methods using a fraction of their computing power.
The scientists said GraphCast outperformed the ECMWF model in more than 90 per cent of over 1,300 test areas and in over 99 per cent of weather variables when it came to the Earth’s troposphere – the lowest part of the atmosphere, where most weather happens.
“Weather prediction is one of the most challenging problems that humanity has been working on for a long, long time,” said Pushmeet Kohli, the vice president of research at Google DeepMind. “And if you look at what has happened in the last few years with climate change, this is an incredibly important problem.”
Weather forecasting is vital in helping to prepare for extreme weather events such as floods and extreme temperatures. For this reason, accurate reporting can save millions of lives.
One such example was Hurricane Lee. DeepMind’s AI predicted that the hurricane would make landfall in Nova Scotia nine days before it happened, giving people three more days to prepare for its arrival than the six traditional methods would have provided.
GraphCast first trained the model using physical models’ estimates of past global weather from 1979 to 2017. This allowed the algorithm to identify links between weather variables such as air pressure, wind, temperature and humidity.
“Once trained, GraphCast is tremendously cheap to operate,” said Matthew Chantry, a machine-learning coordinator at ECMWF. “We might be talking about 1,000 times cheaper in terms of energy consumption. That is a miraculous improvement.”
However, researchers said they anticipate it will be another two to five years before people can use machine-learning forecasting to make real-world decisions.
The researchers’ findings were published in the journal Science.
