The GenCast model has potential socio-economic benefits as it could help mitigate the consequences of severe weather by assisting in better preparation for adverse conditions. It could also aid in renewable energy planning, such as improved wind-power forecasting. The model is cost-effective, with a single Google Cloud TPU v5 able to produce one 15-day forecast in GenCast’s ensemble in just 8 minutes. DeepMind has released the GenCast model code and weights to aid research and development in the weather and climate community.
Key takeaways:
- Google DeepMind researchers have developed a machine learning model, GenCast, that can provide improved 15-day weather forecasts using minimal compute resources.
- GenCast learns directly from historical weather data and can learn more complex relationships and dynamics directly from the data, outperforming traditional models.
- The model has been shown to outperform the top operational ensemble NWP model, ENS, produced by the European Centre for Medium-Range Weather Forecasts, on 97.2 percent of the evaluated targets.
- DeepMind has released the GenCast model code and weights to help accelerate research and development for the weather and climate community, with the release of real-time and historical forecasts from GenCast and prior models planned.