A new machine learning-driven model, called GenCast, has been developed for ensemble-based forecasting, generating probability-based projections for extreme weather events and other meteorological phenomena. Unlike previous deterministic models, GenCast uses multiple initial conditions to produce more accurate predictions.
According to Google DeepMind researchers, the model shows skill in anticipating extreme events outside its training period. However, it will not replace human forecasters but rather serve as an additional tool in their toolkit for predicting day-to-day weather and extreme events.
A study published in Nature found that GenCast has greater skill than a 97.2% reliable ensemble run by the European Center for Medium-Range Weather Forecasts (ECMWF) on most forecast metrics, including predicting tropical cyclones and wind power output.
GenCast’s machine learning approach uses weather data from 1979 to 2018 for training purposes and can generate an ensemble forecast in just 8 minutes. However, some outside meteorologists have raised concerns that the model lacks insights to be a complete forecast model.
Experts acknowledge that AI-driven models like GenCast offer improved forecasting accuracy but caution against moving too fast in adopting the technology. They emphasize the importance of building trust in AI by showcasing its benefits and ensuring its responsible development and deployment.
Source: https://www.axios.com/2024/12/04/google-ai-weather-model-more-reliable