GraphCast from DeepMind researchers outperformed a traditional European weather forecasting system by 99% over 12,000 measurements.
The AI model runs from a desktop computer and makes accurate forecasts in just a few minutes – while the most powerful traditional weather instruments take hours to run, the Nature paper writes.
“GraphCast is currently leading the race among artificial intelligence models,” says Aditya Grover of the University of California, Los Angeles.
Currently, the world's weather forecasting uses what is known as Numerical Weather Prediction (NWP), an approach that uses mathematical models and data from buoys, satellites and weather stations. Calculations show quite accurately how heat, air and water vapor move in the atmosphere, but such instruments are quite expensive and energy-intensive.
Alternative AI tools have already been developed by several well-known technology companies, including DeepMind, computer chip maker NVIDIA, Chinese technology company Huawei, and a number of startups such as Berkeley, California-based Atmo. Artificial intelligence runs 1,000 to 10,000 times faster than conventional NWP models, leaving more time to interpret and communicate predictions.
Huawei's Pangu weather model is the most powerful competitor to the standard NWP system of the European Center for Medium-Range Weather Forecasts (ECMWF) in Reading, UK, which provides the world's best weather forecasts up to 15 days in advance. At the same time, both tools already appear to have outperformed DeepMind's GraphCast, which was trained on weather data from 1979 to 2017 so it could learn relationships between weather variables such as barometric pressure, wind, temperature and humidity.
DeepMind found that GraphCast could also use global weather forecasts from 2018 to make forecasts 10 days ahead in less than a minute, and they were more accurate than the High Resolution Forecast System (HRES), one of the versions of NWP used to forecast the right hours.
Vacancies
Design Lead Spribe
Marketing Director League Design Agency, Kiev
Junior SEO specialist Zeeks
“In the troposphere, GraphCast outperforms HRES by more than 99% in the 12,000 measurements we took,” says Remy Lam of DeepMind in London.
At all atmospheric levels, GraphCast outperformed HRES in 90% of its predictions. The model was also effective at identifying extreme weather events such as tropical cyclones, extreme cold or heatwaves—in one particular example, the tool anticipated Hurricane Lee's approach to Long Island 10 days before it occurred, while traditional weather forecasting technologies which meteorologists used at the time lagged behind.
Compared to the Huawei model, GraphCast was the best in 99% of predictions.
C++ course for GameDevData analyst. A comprehensive course on how to learn to code in C++ from scratch, learn the necessary skills to master the required programming. Find out about the course
At the same time, AI models can improve certain types of weather forecasting that standard tools can't handle—for example, predicting the amount of precipitation that will fall on the ground within a few hours.
“Standard physics models are still needed to produce global weather estimates that are initially used to train AI models,” the researchers say. “We expect it will be another two to five years before people can use machine learning-based predictions to make decisions in the real world.”
GraphCast, or at least the core AI algorithm that powers the predictions, may soon make its way to more mainstream services. According to Wired, Google may already be exploring how to integrate the model into its products.