Google has unveiled GenCast, a cutting-edge AI-driven weather forecasting model, in a paper published in Nature. The model employs a diffusion-based approach, tailored for Earth’s geometry, to offer highly accurate weather predictions.
GenCast has been trained on over 40 years of historical data from ECMWF archives, which enables it to outperform traditional systems, such as ECMWF’s ENS, particularly for longer-range forecasts of up to 25 days.
In rigorous testing, Google ran 1,320 forecasts from 2019 using GenCast, which had been trained on data up until 2018. The results were impressive, with GenCast outperforming ENS in 97.2% of cases, and achieving up to 99.8% accuracy for predictions beyond 36 hours.
For example, when forecasting the path of Typhoon Hagibis in 2019, GenCast was able to refine its predictions as the storm neared landfall, showcasing its potential in improving disaster preparedness and response.
The applications of GenCast go beyond disaster management. It is also highly effective in predicting wind speeds near wind farms and weather conditions at solar farms, providing valuable insights for renewable energy planning.
The model generates more than 50 different predictions with varying probabilities, making it a comprehensive forecasting tool. Additionally, GenCast can generate a 15-day forecast in just eight minutes on Google Cloud TPU v5, significantly reducing processing time compared to traditional supercomputing methods.
In a step toward global collaboration, Google has made GenCast’s code and weights publicly available. This move allows weather agencies and researchers worldwide to refine the model and contribute to the advancement of meteorological science.
With its precision, efficiency, and open-access approach, GenCast marks a significant leap forward in weather forecasting technology.
[embedpost slug="/register-your-vpn-using-mobile-numbers-heres-how/"]
