Despite the promising results, the AI model has limitations. It tends to underestimate the strength of significant events like Category 5 storms and falls short of the ECMWF's predictions for conditions in the stratosphere. The model also doesn't provide ensemble forecasts, which detail multiple potential outcomes for a weather system. However, Google is exploring how to add GraphCast into its products and improve its performance for certain weather conditions.
Key takeaways:
- Google's DeepMind AI unit has developed an AI-based weather model, called GraphCast, that accurately predicted the landfall of Hurricane Lee in September, outperforming traditional forecasts.
- The AI model can generate forecasts in under a minute on a laptop, compared to conventional models that require a supercomputer and can take hours to compute.
- Despite its success, the AI model has limitations, including a tendency to underestimate the strength of major events like Category 5 storms and difficulty in providing ensemble forecasts, which detail multiple potential outcomes for a weather system.
- The European Centre for Medium-Range Weather Forecasting (ECMWF) is developing its own AI weather forecasting model, inspired by GraphCast, and plans to launch AI-powered forecasts in the next year or two.