loading page

A Four-Dimensional Variational Constrained Neural Network-based Data Assimilation Method
  • +5
  • Wuxin Wang,
  • Kaijun Ren,
  • Boheng Duan,
  • Junxing Zhu,
  • Xiaoyong Li,
  • Weicheng Ni,
  • Jingze Lu,
  • Taikang Yuan
Wuxin Wang
College of Computer Science and Technology, National University of Defense Technology, College of Meteorology and Oceanography, National University of Defense Technology
Author Profile
Kaijun Ren
College of Computer Science and Technology, National University of Defense Technology, College of Meteorology and Oceanography, National University of Defense Technology

Corresponding Author:

Boheng Duan
College of Meteorology and Oceanography, National University of Defense Technology

Corresponding Author:

Junxing Zhu
College of Meteorology and Oceanography, National University of Defense Technology
Xiaoyong Li
College of Meteorology and Oceanography, National University of Defense Technology
Weicheng Ni
College of Computer Science and Technology, National University of Defense Technology, College of Meteorology and Oceanography, National University of Defense Technology
Jingze Lu
College of Computer Science and Technology, National University of Defense Technology, College of Meteorology and Oceanography, National University of Defense Technology
Taikang Yuan
College of Meteorology and Oceanography, National University of Defense Technology

Abstract

• A physics-informed neural network trained without ground truths can provide accurate initial fields for numerical prediction. • The system's kinetic features are embedded into the model through our four-dimensional variational form loss function. • We show on Lorenz96 that the proposed method can be used directly for accurate data assimilation at a low computational cost.
21 Dec 2023Submitted to ESS Open Archive
21 Dec 2023Published in ESS Open Archive
Jan 2024Published in Journal of Advances in Modeling Earth Systems volume 16 issue 1. 10.1029/2023MS003687