🌧 Know About Rain 2 Weeks in Advance: How DeepMind's New Weather Model Works
The GenCast model, consisting of an ensemble of AI tools (around 50 different algorithms), predicts the weather up to 15 days ahead.
The model's results are more accurate than those of the European Centre for Medium-Range Weather Forecasts (ECMWF) in 97.2% of tests and in 99.8% of cases for forecasts beyond 1.5 days. ECMWF data is widely used by weather forecasting services and international organizations.
🌍 How GenCast Was Trained
GenCast is a diffusion model, the same type that powers popular image, video, and music generators. It is adapted to Earth's spherical geometry and uses the latest weather conditions as input data.
The model was trained on 40 years of historical weather data from the ECMWF archive, which includes detailed descriptions of temperature, wind speed, and pressure at various altitudes.
☁️ Advantages of GenCast
DeepMind's development is not only more accurate but also significantly faster than traditional systems. GenCast generates a 15-day forecast in just 8 minutes, while conventional algorithms running on supercomputers take several hours to produce similar results.
GenCast also excels at predicting extreme weather changes, including the trajectories of tropical storms and hurricanes. The system successfully forecasted a typhoon's movement off Japan's coast more than 7 days in advance. Such lead times allow governments to take early safety measures, such as evacuations and stockpiling supplies.
#️⃣ DeepMind has also promised to release the model's code, weights, and forecasts to support a broader community of weather forecasters.
More on the topic:
💨 AI Predicts Weather and Air Pollution
