The European Centre for Medium-range Weather Forecast (ECMWF) hailed GraphCast as a significant advancement in the industry. Matthew Chantry, ECMWF machine learning specialist, suggested that once trained on a wider variety of historical data, models like GraphCast could be much cheaper than current methods that rely on powerful supercomputers. He estimated a potential 1,000 times reduction in energy consumption, calling it a "miraculous improvement".
Key takeaways:
- Deepmind's GraphCast AI can produce an accurate 10-day weather forecast in under a minute, outperforming traditional weather forecasters for the first time.
- The AI model was able to predict Hurricane Lee's landfall three days before traditional forecasters.
- Despite its success, the model does have some limitations when dealing with the uncertainty inherent in longer-term weather forecasts.
- Once trained on a wider variety of historical data, models like GraphCast could be much cheaper than current weather forecasting methods, potentially reducing energy consumption by 1,000 times.