Researchers at Google DeepMind have developed GenCast, a new AI tool that predicts weather more accurately than current models, as highlighted in a recent study published in Nature.
GenCast distinguishes itself from traditional forecasting methods, which have relied on mathematical equations since the 1920s. Instead, it uses machine learning techniques, trained on 40 years of weather data, to generate forecasts for 15 days across 80 locations in just 8 minutes.
In tests against the European Centre for Medium-Range Weather Forecasts, GenCast outperformed traditional models in 97.2% of 1,320 predictions. Its capabilities extend to forecasting extreme weather events and predicting wind energy production, making it a significant advancement in climate change mitigation and renewable energy planning.
Despite its success, GenCast operates at a resolution of 0.25 degrees, which is less precise than some existing models. However, it introduces a diffusion model that can generate probabilistic forecasts, enhancing its ability to predict extreme weather scenarios.