Computing the Ensemble Spread From Deterministic Weather Predictions Using Conditional Generative Adversarial Networks

Ensemble prediction systems are an invaluable tool for weather forecasting. Practically, ensemble predictions obtained by running several perturbations of the deterministic control forecast. However, is associated with a high computational cost and often involves statistical post-processing steps to improve its quality. Here we propose use deep-learning-based algorithms learn properties system, spread, given only Thus, once trained, costly system will not be needed anymore obtain future forecasts, can derived from single We adapt classical pix2pix architecture three-dimensional model also experiment shared latent space encoder-decoder model, train them against years operational (ensemble) forecasts 500 hPa geopotential height. The results demonstrate that trained models indeed allow obtaining highly accurate spread forecast only.