SEAToday.com, Jakarta - Google has unveiled GenCast, an AI-powered high-resolution ensemble model designed to revolutionize weather forecasting. Announced on (12/6/2024), this cutting-edge model promises unprecedented accuracy and efficiency in predicting weather patterns.

GenCast outperforms current systems like ECMWF's ENS, boasting a 97.2% accuracy rate in 25-day forecasts and up to 99.8% for 36-hour projections. 

Built on four decades of historical weather data, it uses advanced diffusion models fine-tuned for Earth's unique geometry. "GenCast transforms how we prepare for weather-related challenges," Google remarked.

A key demonstration included mapping Typhoon Hagibis' 2019 trajectory in Japan. GenCast's predictions closely matched actual events, offering valuable lead time for disaster preparedness. Google highlighted this as a breakthrough for local governments facing severe weather.

Beyond storms, GenCast aids renewable energy by forecasting wind speeds near turbines and weather over solar fields. Its ensemble approach generates over 50 probabilistic predictions within minutes on Google Cloud TPU v5, surpassing traditional models requiring hours on supercomputers.

Released as an open model, GenCast's code and weights are accessible to scientists worldwide. Google plans ongoing collaborations with meteorological agencies to further enhance global weather forecasting.

Writer: Fariz Ahmad Luthfi