SUBGRID-SCALE SURROGATE MODELING OF IDEALIZED ATMOSPHERIC FLOWS: A DEEP LEARNED APPROACH USING HIGH-RESOLUTION SIMULATION DATA

Muralikrishnan Gopalakrishnan Meena; Matthew R. Norman; David M. Hall

SUBGRID-SCALE SURROGATE MODELING OF IDEALIZED ATMOSPHERIC FLOWS: A DEEP LEARNED APPROACH USING HIGH-RESOLUTION SIMULATION DATA

Muralikrishnan Gopalakrishnan Meena, Matthew R. Norman, David M. Hall

Advanced Computing for Life Sciences and

Research output: Contribution to conference › Paper › peer-review

Abstract

We introduce a deep learned subgrid-scale surrogate model for dry, stratified idealized atmospheric flows from high-resolution simulation data. Deep neural networks (NNs) are used to model the full state differences between a coarse resolution simulation and a high-resolution simulation, run simultaneously with the coarse resolution simulation forced by the high-resolution simulation, hence capturing both dissipative and anti-dissipative effects. The NN model is able to accurately capture the state differences in a priori tests outside the training regime. In a posteriori tests intended for production use, the NN coupled coarse simulation is accurate compared to the high-resolution simulation over a finite period in time. With the accumulation of the errors, the NN-coupled simulation becomes computationally unstable after a while. These surrogate models further pave the way for formulating stable, complex, physics-based NN models which are driven by traditional subgrid-scale turbulence closure models.

Original language	English
State	Published - 2022
Event	12th International Symposium on Turbulence and Shear Flow Phenomena, TSFP 2022 - Osaka, Virtual, Japan Duration: Jul 19 2022 → Jul 22 2022

Conference

Conference	12th International Symposium on Turbulence and Shear Flow Phenomena, TSFP 2022
Country/Territory	Japan
City	Osaka, Virtual
Period	07/19/22 → 07/22/22

Funding

This research was supported by the Exascale Computing Project (17-SC-20-SC), a collaborative effort of the U.S. Department of Energy Office of Science and the National Nuclear Security Administration. This research used resources of the Oak Ridge Leadership Computing Facility at the Oak Ridge National Laboratory, which is supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC05-00OR22725.

Funders	Funder number
U.S. Department of Energy	DE-AC05-00OR22725
U.S. Department of Energy
Office of Science
National Nuclear Security Administration

Cite this

@conference{297f61dc9b50454c9c25c0837c18eef9,

title = "SUBGRID-SCALE SURROGATE MODELING OF IDEALIZED ATMOSPHERIC FLOWS: A DEEP LEARNED APPROACH USING HIGH-RESOLUTION SIMULATION DATA",

abstract = "We introduce a deep learned subgrid-scale surrogate model for dry, stratified idealized atmospheric flows from high-resolution simulation data. Deep neural networks (NNs) are used to model the full state differences between a coarse resolution simulation and a high-resolution simulation, run simultaneously with the coarse resolution simulation forced by the high-resolution simulation, hence capturing both dissipative and anti-dissipative effects. The NN model is able to accurately capture the state differences in a priori tests outside the training regime. In a posteriori tests intended for production use, the NN coupled coarse simulation is accurate compared to the high-resolution simulation over a finite period in time. With the accumulation of the errors, the NN-coupled simulation becomes computationally unstable after a while. These surrogate models further pave the way for formulating stable, complex, physics-based NN models which are driven by traditional subgrid-scale turbulence closure models.",

author = "Meena, {Muralikrishnan Gopalakrishnan} and Norman, {Matthew R.} and Hall, {David M.}",

note = "Publisher Copyright: {\textcopyright} 2022 12th International Symposium on Turbulence and Shear Flow Phenomena, TSFP 2022. All rights reserved.; 12th International Symposium on Turbulence and Shear Flow Phenomena, TSFP 2022 ; Conference date: 19-07-2022 Through 22-07-2022",

year = "2022",

language = "English",

}

SUBGRID-SCALE SURROGATE MODELING OF IDEALIZED ATMOSPHERIC FLOWS: A DEEP LEARNED APPROACH USING HIGH-RESOLUTION SIMULATION DATA. / Meena, Muralikrishnan Gopalakrishnan ; Norman, Matthew R.; Hall, David M.
2022. Paper presented at 12th International Symposium on Turbulence and Shear Flow Phenomena, TSFP 2022, Osaka, Virtual, Japan.

Research output: Contribution to conference › Paper › peer-review

TY - CONF

T1 - SUBGRID-SCALE SURROGATE MODELING OF IDEALIZED ATMOSPHERIC FLOWS

T2 - 12th International Symposium on Turbulence and Shear Flow Phenomena, TSFP 2022

AU - Meena, Muralikrishnan Gopalakrishnan

AU - Norman, Matthew R.

AU - Hall, David M.

PY - 2022

Y1 - 2022

N2 - We introduce a deep learned subgrid-scale surrogate model for dry, stratified idealized atmospheric flows from high-resolution simulation data. Deep neural networks (NNs) are used to model the full state differences between a coarse resolution simulation and a high-resolution simulation, run simultaneously with the coarse resolution simulation forced by the high-resolution simulation, hence capturing both dissipative and anti-dissipative effects. The NN model is able to accurately capture the state differences in a priori tests outside the training regime. In a posteriori tests intended for production use, the NN coupled coarse simulation is accurate compared to the high-resolution simulation over a finite period in time. With the accumulation of the errors, the NN-coupled simulation becomes computationally unstable after a while. These surrogate models further pave the way for formulating stable, complex, physics-based NN models which are driven by traditional subgrid-scale turbulence closure models.

AB - We introduce a deep learned subgrid-scale surrogate model for dry, stratified idealized atmospheric flows from high-resolution simulation data. Deep neural networks (NNs) are used to model the full state differences between a coarse resolution simulation and a high-resolution simulation, run simultaneously with the coarse resolution simulation forced by the high-resolution simulation, hence capturing both dissipative and anti-dissipative effects. The NN model is able to accurately capture the state differences in a priori tests outside the training regime. In a posteriori tests intended for production use, the NN coupled coarse simulation is accurate compared to the high-resolution simulation over a finite period in time. With the accumulation of the errors, the NN-coupled simulation becomes computationally unstable after a while. These surrogate models further pave the way for formulating stable, complex, physics-based NN models which are driven by traditional subgrid-scale turbulence closure models.

UR - http://www.scopus.com/inward/record.url?scp=85143764529&partnerID=8YFLogxK

M3 - Paper

AN - SCOPUS:85143764529

Y2 - 19 July 2022 through 22 July 2022

ER -

SUBGRID-SCALE SURROGATE MODELING OF IDEALIZED ATMOSPHERIC FLOWS: A DEEP LEARNED APPROACH USING HIGH-RESOLUTION SIMULATION DATA

Abstract

Conference

Funding

Other files and links

Fingerprint

Cite this