Speaker
Description
This study explores the adaptation of state-of-the-art deep learning architectures for video frame prediction in the context of weather and climate applications. As a case study, we attempt to predict surface temperature fields over Europe for up to 24 hours based on meteorological analyses. Initial results have been achieved with a PredNet and a GAN architecture. To facilitate the massive data processing and testing of different deep learning architectures, we have developed a containerized workflow for the full life-cycle of the application, which consists of data extraction, data pre-processing, training, post-processing, and visualization of results. The containerized workflow allows for porting the workflow to different machines, including the massively parallel JUWELS supercomputer at Jülich Supercomputing Center. The presentation will focus on the workflow aspects but also discuss a few preliminary results obtained with both network architectures.