Ensemble Kalman Filter based on the Image Structures

Béréziat, Dominique; Herlin, Isabelle; Lepoittevin, Yann

doi:10.5220/0006096201400150

Cited by 1 publication

(2 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As proved in Lepoittevin and Herlin (2016), these terms can be directly included in the background error using a particular matrix such that , where and are the parameters to be tuned. Such regularization is a classical optical flow penalty (Horn and Schunck, 1981) and can be used for the sea-surface circulation estimation (Béréziat, 2000, Béréziat and Herlin, 2014).…”

Section: Case Studymentioning

confidence: 99%

“…In this work, we propose a hybrid methodology bridging deep prior and variational data assimilation and we test it in a twin experiment. The algorithm is evaluated on an ocean-like motion estimation task requiring regularization, then compared to adapted data assimilation algorithms (Béréziat and Herlin, 2014, 2018). All algorithms are implemented using tools from the deep learning community.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Deep prior in variational assimilation to estimate an ocean circulation without explicit regularization

Filoche¹,

Béréziat²,

Charantonis

2023

Environ. Data Science

View full text Add to dashboard Cite

Many applications in geosciences require solving inverse problems to estimate the state of a physical system. Data assimilation provides a strong framework to do so when the system is partially observed and its underlying dynamics are known to some extent. In the variational flavor, it can be seen as an optimal control problem where initial conditions are the control parameters. Such problems are often ill-posed, regularization may be needed using explicit prior knowledge to enforce a satisfying solution. In this work, we propose to use a deep prior, a neural architecture that generates potential solutions and acts as implicit regularization. The architecture is trained in a fully-unsupervised manner using the variational data assimilation cost so that gradients are backpropagated through the dynamical model and then through the neural network. To demonstrate its use, we set a twin experiment using a shallow-water toy model, where we test various variational assimilation algorithms on an ocean-like circulation estimation.

show abstract