A comparison of combined data assimilation and machine learning methods for offline and online model error correction

Farchi, Alban; Bocquet, Marc; Laloyaux, Patrick; Bonavita, Massimo; Malartic, Quentin

doi:10.1016/j.jocs.2021.101468

Cited by 32 publications

(44 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…(2021) and Farchi et al. (2021). For the RNN used here, it is straightforward to perform continued online training using low rank matrix updates to update the linear readout operator W out .…”

Section: Discussionmentioning

confidence: 95%

Integrating Recurrent Neural Networks With Data Assimilation for Scalable Data‐Driven State Estimation

Penny

Smith

Chen

et al. 2022

J Adv Model Earth Syst

View full text Add to dashboard Cite

Data assimilation (DA) is integrated with machine learning in order to perform entirely data‐driven online state estimation. To achieve this, recurrent neural networks (RNNs) are implemented as pretrained surrogate models to replace key components of the DA cycle in numerical weather prediction (NWP), including the conventional numerical forecast model, the forecast error covariance matrix, and the tangent linear and adjoint models. It is shown how these RNNs can be initialized using DA methods to directly update the hidden/reservoir state with observations of the target system. The results indicate that these techniques can be applied to estimate the state of a system for the repeated initialization of short‐term forecasts, even in the absence of a traditional numerical forecast model. Further, it is demonstrated how these integrated RNN‐DA methods can scale to higher dimensions by applying domain localization and parallelization, providing a path for practical applications in NWP.

show abstract

Section: Discussionmentioning

confidence: 95%

Integrating Recurrent Neural Networks With Data Assimilation for Scalable Data‐Driven State Estimation

Penny

Smith

Chen

et al. 2022

J Adv Model Earth Syst

View full text Add to dashboard Cite

show abstract

“…The typical applications we have in mind are at that ML/DA frontier where part or the whole model needs to be learned. Moreover, following Bocquet et al (2021), we aim to address the difficult objective of learning state and parameters on the fly, that is, online as observations are acquired, using sequential DA techniques such as the ensemble Kalman filter (EnKF: Evensen, 2009) as an alternative to the variational methods that are more common for parameter and ML problems (Farchi et al, 2021a).…”

Section: Parameter Estimation and Data-driven Techniques For The Geos...mentioning

confidence: 99%

State, global, and local parameter estimation using local ensemble Kalman filters: Applications to online machine learning of chaotic dynamics

Malartic

Farchi

Bocquet

2022

Quart J Royal Meteoro Soc

Self Cite

View full text Add to dashboard Cite

In a recent methodological article, we showed how to learn chaotic dynamics along with the state trajectory from sequentially acquired observations, using local ensemble Kalman filters. Here, we investigate more systematically the possibility of using a local ensemble Kalman filter with either covariance localisation or local domains, in order to retrieve the state and a mix of key global and local parameters. Global parameters are meant to represent the surrogate dynamical core, for instance through a neural network, which is reminiscent of data-driven machine learning of dynamics, while the local parameters typically stand for the forcings of the model. Aiming at joint state and parameter estimation, a family of algorithms for covariance and local domain localisation is proposed. In particular, we show how to update global parameters rigorously using a local-domain ensemble Kalman filter (EnKF) such as the local ensemble transform Kalman filter (LETKF), an inherently local method. The approach is tested with success on the 40-variable Lorenz model using several of the local EnKF flavors. A two-dimensional illustration based on a multilayer Lorenz model is finally provided. It uses radiance-like nonlocal observations. It features both local domains and covariance localisation, in order to learn the chaotic dynamics and the local forcings. This article addresses more generally the key question of online estimation of both global and local model parameters.

show abstract

“…Weak‐constraint 4D‐Var can be seen as a specific ML algorithm that learns the model error by estimating the parameters in the forcing vector (Farchi, Bocquet, et al., 2021). However, there are several conceptual differences with the ML approach described in Section 4.…”

Section: Weak‐constraint 4d‐varmentioning

confidence: 99%

Deep Learning to Estimate Model Biases in an Operational NWP Assimilation System

Laloyaux

Kurth

Dueben

et al. 2022

J Adv Model Earth Syst

Self Cite

View full text Add to dashboard Cite

Machine learning (ML) has made rapid progress in many domains including natural language processing, computer vision, autonomous driving, healthcare and finance (Goodfellow et al., 2016). ML applications can be very complex, and neural networks (NN) can consist of millions to billions of trainable parameters, large numbers of layers, and specialized architectures. In recent years, the weather and climate modeling community has started to explore ML techniques with many applications in numerical weather predictions (NWP;Dueben et al., 2021). In general, these applications can be divided into three groups: methods that improve computational efficiency, methods that improve the quality of the prediction system, and methods that help improve our understanding of the Earth system, for example, via unsupervised learning and causal discovery. This paper belongs to the group that aims to improve prediction quality. In particular, we will use deep learning to learn the systematic error of weather forecast models. Attempts to use DL techniques to estimate and correct for model errors have already been documented in the geophysical literature. For example, Watson (2019) uses an Artificial Neural Network to estimate model error tendencies in the Lorenz-96 system. Predicting the error via deep learning is appealing,

show abstract

A comparison of combined data assimilation and machine learning methods for offline and online model error correction

Cited by 32 publications

References 34 publications

Integrating Recurrent Neural Networks With Data Assimilation for Scalable Data‐Driven State Estimation

Integrating Recurrent Neural Networks With Data Assimilation for Scalable Data‐Driven State Estimation

State, global, and local parameter estimation using local ensemble Kalman filters: Applications to online machine learning of chaotic dynamics

Deep Learning to Estimate Model Biases in an Operational NWP Assimilation System

Contact Info

Product

Resources

About