Presentation | NG21A: Advances in Data Assimilation, Data Fusion, Machine Learning, Predictability, and Uncertainty Quantification in the Geosciences I Oral
Author(s): Bohan Chen, California Institute of Technology (First Author, Presenting Author) Eviatar Bach, Laboratoire de Météorologie Dynamique ENS Edoardo Calvello, California Institute of Technology Ricardo Baptista, California Institute of Technology Andrew Stuart, California Institute of Technology
We created a new machine-learning tool to track of complex, changing systems—like weather patterns—more accurately. When we tested it on some widely used models, it predicted the true state of the system more closely than other leading methods. Our key innovation is the ability to take the entire ensemble as input, without relying on a Gaussian assumption, and to train at a single ensemble size while deploying seamlessly across ensembles of different sizes.
