Deep Learning for Weather Prediction and Ensemble Post-Processing
Presenter
DescriptionRecent advances in deep learning have shown great promise in weather and climate tasks. However, the neural network architectures and training infrastructure are often adapted from "off-the-shelf" approaches for tasks such as image classification. Such approaches need to be scaled and tuned for these new tasks, and understanding the importance of different components is critical. We will present a suite of successes, failures, and the reasons behind them when applying diverse architectures, such as U-Nets, attention, and vision transformers, to two important weather tasks: post-processing ensembles from numerical weather prediction systems and medium-term weather prediction. We will also discuss key aspects for successfully training such models at scale, including optimized I/O ingestion and distributed training. By combining insights from both areas, we are able to demonstrate improved results on these tasks and provide guidance for the community.
TimeMonday, June 2717:30 - 18:00 CEST
LocationSydney Room
Event Type
Minisymposium
Climate, Weather and Earth Sciences
Computer Science and Applied Mathematics
Engineering
Physics
Author
