Deep learning for post-processing ensemble weather forecasts

Quantifying uncertainty in weather forecasts is critical, especially for predicting extreme events. This typically accomplished with ensemble prediction systems, which consist of many perturbed numerical simulations, or trajectories, run parallel. These systems are associated a high computational cost and often involve statistical post-processing steps to inexpensively improve their raw qualities. We propose mixed model that uses only subset the original trajectories combined step using deep neural networks. enable account non-linear relationships not captured by current models methods. Applied global data, our achieve relative improvement forecast skill (CRPS) over 14%. Furthermore, we demonstrate larger events on select case studies. also show can use fewer comparable results full ensemble. By costs an system be reduced, allowing it at higher resolution produce more accurate forecasts. article part theme issue ‘Machine learning climate modelling’.