They are stepping on the heels of meteorologists: Google DeepMind has developed an AI model that accurately determines the weather for the next 10 days
Google DeepMind has created a new artificial intelligence model GraphCast that can generate ten-day weather forecasts in less than a minute. According to the developers, it is as accurate as traditional numerical forecasting methods.
GraphCast is a graph neural network trained on 39 years of weather data. It analyzes the relationships of various weather variables and builds a forecast on this basis.
Unlike traditional approaches, the DeepMind model uses data from satellites, radars, and weather stations. At the same time, its training requires significant computing resources.
According to testing, GraphCast outperforms the ECMWF benchmark system in terms of forecast accuracy for most variables and forecast horizons. The model has proven to be particularly good at analyzing the troposphere.
However, the DeepMind model has drawbacks. Its forecasts can be incomplete and complicated at high altitudes. According to the developers, AI cannot yet completely replace traditional numerical forecasting methods.
Google DeepMind has published the GraphCast code on GitHub.
