Fast data assimilation (FDA): Data assimilation by machine learning for faster optimize model state

Pin Wu; Xuting Chang; Wenyan Yuan; Junwu Sun; Wenjie Zhang; Rossella Arcucci; Yi-Ke GUO

doi:10.1016/j.jocs.2021.101323

Fast data assimilation (FDA): Data assimilation by machine learning for faster optimize model state

Pin Wu^*, Xuting Chang, Wenyan Yuan, Junwu Sun, Wenjie Zhang, Rossella Arcucci, Yi-Ke GUO

^*Corresponding author for this work

Office of the Vice-President (Research and Development)

Research output: Contribution to journal › Journal article › peer-review

20 Citations (Scopus)

Abstract

Data assimilation (DA) can provide the more accurate initial state for numerical forecasting models. But traditional DA algorithms has the problem of long calculation time. This paper proposes fast data assimilation (FDA) based on machine learning. For training model, FDA uses 4DVAR, iForest, MLP, and also includes a modified model that does not require observations. This paper applies FDA in the Lorenz63 dynamical system. The experimental results show that the single analysis time of FDA is almost 524 times faster than 4DVAR. FDA greatly reduces the time of the DA process.

Original language	English
Article number	101323
Journal	Journal of Computational Science
Volume	51
Early online date	10 Feb 2021
DOIs	https://doi.org/10.1016/j.jocs.2021.101323
Publication status	Published - Apr 2021

User-Defined Keywords

Data assimilation
Four-Dimensional Variational Assimilation
Machine learning
Multi-layer perceptron
Numerical forecast

Access to Document

10.1016/j.jocs.2021.101323

Cite this

@article{da35c4dcf0914c568aae3a0f17b7f336,

title = "Fast data assimilation (FDA): Data assimilation by machine learning for faster optimize model state",

abstract = "Data assimilation (DA) can provide the more accurate initial state for numerical forecasting models. But traditional DA algorithms has the problem of long calculation time. This paper proposes fast data assimilation (FDA) based on machine learning. For training model, FDA uses 4DVAR, iForest, MLP, and also includes a modified model that does not require observations. This paper applies FDA in the Lorenz63 dynamical system. The experimental results show that the single analysis time of FDA is almost 524 times faster than 4DVAR. FDA greatly reduces the time of the DA process.",

keywords = "Data assimilation, Four-Dimensional Variational Assimilation, Machine learning, Multi-layer perceptron, Numerical forecast",

author = "Pin Wu and Xuting Chang and Wenyan Yuan and Junwu Sun and Wenjie Zhang and Rossella Arcucci and Yi-Ke GUO",

note = "Funding Information: This work is supported by State Key Laboratory of Aerodynamics, China , China Aerodynamics Research and Development Center under Grant SKLA20180303 , Natural Science Foundation of Shanghai, China under Grant 19ZR1417700 , and Transforming Systems through Partnership, Newton Fund under Grant TSPC1086 . ",

year = "2021",

month = apr,

doi = "10.1016/j.jocs.2021.101323",

language = "English",

volume = "51",

journal = "Journal of Computational Science",

issn = "1877-7503",

publisher = "Elsevier B.V.",

}

TY - JOUR

T1 - Fast data assimilation (FDA)

T2 - Data assimilation by machine learning for faster optimize model state

AU - Wu, Pin

AU - Chang, Xuting

AU - Yuan, Wenyan

AU - Sun, Junwu

AU - Zhang, Wenjie

AU - Arcucci, Rossella

AU - GUO, Yi-Ke

N1 - Funding Information: This work is supported by State Key Laboratory of Aerodynamics, China , China Aerodynamics Research and Development Center under Grant SKLA20180303 , Natural Science Foundation of Shanghai, China under Grant 19ZR1417700 , and Transforming Systems through Partnership, Newton Fund under Grant TSPC1086 .

PY - 2021/4

Y1 - 2021/4

N2 - Data assimilation (DA) can provide the more accurate initial state for numerical forecasting models. But traditional DA algorithms has the problem of long calculation time. This paper proposes fast data assimilation (FDA) based on machine learning. For training model, FDA uses 4DVAR, iForest, MLP, and also includes a modified model that does not require observations. This paper applies FDA in the Lorenz63 dynamical system. The experimental results show that the single analysis time of FDA is almost 524 times faster than 4DVAR. FDA greatly reduces the time of the DA process.

AB - Data assimilation (DA) can provide the more accurate initial state for numerical forecasting models. But traditional DA algorithms has the problem of long calculation time. This paper proposes fast data assimilation (FDA) based on machine learning. For training model, FDA uses 4DVAR, iForest, MLP, and also includes a modified model that does not require observations. This paper applies FDA in the Lorenz63 dynamical system. The experimental results show that the single analysis time of FDA is almost 524 times faster than 4DVAR. FDA greatly reduces the time of the DA process.

KW - Data assimilation

KW - Four-Dimensional Variational Assimilation

KW - Machine learning

KW - Multi-layer perceptron

KW - Numerical forecast

UR - http://www.scopus.com/inward/record.url?scp=85101903229&partnerID=8YFLogxK

U2 - 10.1016/j.jocs.2021.101323

DO - 10.1016/j.jocs.2021.101323

M3 - Journal article

AN - SCOPUS:85101903229

SN - 1877-7503

VL - 51

JO - Journal of Computational Science

JF - Journal of Computational Science

M1 - 101323

ER -

Fast data assimilation (FDA): Data assimilation by machine learning for faster optimize model state

Abstract

User-Defined Keywords

Access to Document

Other files and links

Fingerprint

Cite this