Why in the News?
Google DeepMind has unveiled its revolutionary GenCast AI model, designed to predict the weather more accurately and farther in advance than current forecasting tools.
About GenCast:
| What is it? |
- GenCast is an AI-based weather forecasting model developed by Google DeepMind.
- It uses machine learning techniques for more accurate and long-term predictions compared to traditional models.
- Unlike traditional numerical weather prediction (NWP) models, GenCast uses an ensemble of AI-generated forecasts, trained on 40 years of reanalysis data.
- Outperforms traditional tools in predicting extreme weather, tropical cyclones, and wind power production.
|
| How GenCast Works |
- Trained on 40 years of reanalysis data (1979–2019), blending historical data and modern forecasts.
- It is powered by a neural network with 41,162 nodes and 240,000 edges, where nodes process data and edges connect them.
- A diffusion model that refines noisy data in 30 steps to improve forecast accuracy.
- It generates about 50 forecasts at once, providing probabilistic predictions (e.g., likelihood of rain, not exact amounts).
- Generates forecasts in 8 minutes using a single TPU v5 unit, much faster than traditional NWP models, which take hours.
|
| Significance of GenCast |
- Outperforms ECMWF ensemble forecasts on 97.2% of targets, especially for extreme weather predictions.
- Provides longer-term forecasts for up to 15 days, with spatial resolution of 0.25° x 0.25° and 12-hour intervals.
- Offers probabilistic forecasts to help better prepare for extreme weather.
- Faster processing than traditional models, reducing forecast time from hours to minutes.
- Sustainability and scalability allow the model to be expanded to other areas of weather prediction.
- Google collaborates with weather agencies to enhance AI forecasting methods while recognizing the importance of traditional models.
|
