Google DeepMind’s AI has just humiliated human meteorologists

Is AI about to win out over humanity? In any case, GraphCast, Google’s AI DeepMind, has just achieved a feat. The tool was able to predict weather conditions before any other specialist in the field.

Weather conditions on the planet are subject to unpredictable changes. Even the most powerful tools have a little difficulty predicting these events accurately. But GraphCast, the AI from Google Deepmind could solve this problem. Its latest exploits have taken the world by surprise. For example, this tool was able to anticipate a hurricane in Nova Scotia three days before conventional methods.

Exceptional accuracy combined with fast operation

According to specialists, the GraphCast AI is capable of predict weather conditions 10 days in advance. Traditional tools lag far behind this feat. And this conclusion was reached after specific tests. Scientists compared GraphCast with the European model for medium-range weather forecasting (ECMWF). They compared the two innovations over 1300 analysis zones. GraphCast was accurate in 90% of the 1300 given variables.

Google DeepMind’s AI is then a reference tool for predicting weather conditions in the Earth’s troposphere. It will be able to study air temperatures more accurately. AI has also proved its worth in anticipating dangerous phenomena such as hurricanes or extreme heat.

Google Deepmind’s AI focuses on machine learning to achieve this exceptional performance. GraphCast draws its forecasts from extensive weather data. That’s right, the AI uses information from four decadeshence the accuracy of the results. The tool then combines this information with physics-based equations.

“Once trained, GraphCast is extremely inexpensive to use. We could be talking about 1,000 times cheaper in terms of energy consumption” Matthew Chantry, Machine Learning Specialist, ECMWF.

GraphCast maps the planet’s surface with a million grid points. At each of these zones, the AI predicts temperature, wind direction and speed. It also analyzes average sea-level pressure and humidity. The neural network then examines all this data before drawing conclusions. And these calculations take less than a minute.

GraphCast does have a few drawbacks, however. During testing, the tool had some difficulty in predicting classic weather conditions.ues, such as precipitation. Experts then had to combine the classical method with GraphCast to obtain accurate predictions on all fronts.

A matter of competition

In recent years, the field of meteorology has undergone a meteoric evolution. The high-tech giants have all released high-performance tools to improve event prediction.

Huawei’s Pangu-Weather and Nvidia’s FourcastNet have already proved their worth. GraphCast has also contributed to this sector.

“GraphCast is currently leading the race among AI models,” said Aditya, Computer Scientist, University of California.

But despite the competition, these tools can work together to give the best possible rendering.