Core Concepts
Machine learning methods enhance ensemble-based data assimilation, with UNetKF showing superior performance.
Abstract
Introduction to U-Net Kalman Filter (UNetKF) and its application in ensemble data assimilation.
Training U-Nets using EnKF experiments to predict error covariances.
Performance comparison of UNetKF with 3DVar, En3DVar, and EnKF in QG-L model.
Transferability of trained U-Nets to higher-resolution QG-H model for UNetKF experiments.
Evaluation of RMSE sensitivities on DA parameters and computational efficiency considerations.
Future research directions and potential improvements in ML-assisted data assimilation.
Stats
U-Netは、局所アンサンブル誤差共分散を予測するために訓練されます。