Long short-term memory embedded nudging schemes for nonlinear data assimilation of geophysical flows

TL;DR

This paper introduces a novel LSTM-based neural network method for nonlinear data assimilation in geophysical flows, offering a simple, stable, and more accurate alternative to traditional filters, especially with sparse observations.

Contribution

The paper proposes a fully nonintrusive LSTM embedding architecture to estimate nudging terms, improving data assimilation accuracy and efficiency over EKF and EnKF methods.

Findings

LSTM nudging outperforms EKF and EnKF with sparse data

The approach is computationally more efficient

Model retraining is effective via transfer learning

Abstract

Reduced rank nonlinear filters are increasingly utilized in data assimilation of geophysical flows, but often require a set of ensemble forward simulations to estimate forecast covariance. On the other hand, predictor-corrector type nudging approaches are still attractive due to their simplicity of implementation when more complex methods need to be avoided. However, optimal estimate of nudging gain matrix might be cumbersome. In this paper, we put forth a fully nonintrusive recurrent neural network approach based on a long short-term memory (LSTM) embedding architecture to estimate the nudging term, which plays a role not only to force the state trajectories to the observations but also acts as a stabilizer. Furthermore, our approach relies on the power of archival data and the trained model can be retrained effectively due to power of transfer learning in any neural network applications. In order to verify the feasibility of the proposed approach, we perform twin experiments using Lorenz 96 system. Our results demonstrate that the proposed LSTM nudging approach yields more accurate estimates than both extended Kalman filter (EKF) and ensemble Kalman filter (EnKF) when only sparse observations are available. With the availability of emerging AI-friendly and modular hardware technologies and heterogeneous computing platforms, we articulate that our simplistic nudging framework turns out to be computationally more efficient than either the EKF or EnKF approaches.