An Improved Perturbation Pressure Closure for Eddy-Diffusivity Mass-Flux Schemes

He, Jia; Cohen, Yair; Lopez‐Gomez, Ignacio; Jaruga, Anna; Schneider, Tapio

doi:10.1002/essoar.10505084.2

Cited by 5 publications

(6 citation statements)

References 50 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We consider the EDMF scheme discussed in Cohen et al (2020); Lopez-Gomez et al (2020), which is implemented in a single-column model (SCM). Within this SCM, we first seek to learn 16 closure parameters: Five describing turbulent mixing, dissipation, and mixing inhibition by stratification (Lopez-Gomez et al, 2020), three describing the momentum exchange between subdomains (He et al, 2021), seven describing entrainment fluxes between updrafts and the environment (Cohen et al, 2020), and another one defining the surface area fraction occupied by updrafts. In Section 4.4, we substitute the empirical dynamical entrainment closure proposed in Cohen et al (2020) by a neural network, and train the resulting physics-based machine-learning model.…”

Section: Application To An Atmospheric Subgrid-scale Modelmentioning

confidence: 99%

See 1 more Smart Citation

Training Physics‐Based Machine‐Learning Parameterizations With Gradient‐Free Ensemble Kalman Methods

Lopez‐Gomez

Christopoulos

Ervik

et al. 2022

J Adv Model Earth Syst

Self Cite

View full text Add to dashboard Cite

Most machine learning applications in Earth system modeling currently rely on gradientbased supervised learning. This imposes stringent constraints on the nature of the data used for training (typically, residual time tendencies are needed), and it complicates learning about the interactions between machine-learned parameterizations and other components of an Earth system model. Approaching learning about process-based parameterizations as an inverse problem resolves many of these issues, since it allows parameterizations to be trained with partial observations or statistics that directly relate to quantities of interest in long-term climate projections. Here we demonstrate the effectiveness of Kalman inversion methods in treating learning about parameterizations as an inverse problem. We consider two different algorithms: unscented and ensemble Kalman inversion. Both methods involve highly parallelizable forward model evaluations, converge exponentially fast, and do not require gradient computations. In addition, unscented Kalman inversion provides a measure of parameter uncertainty. We illustrate how training parameterizations can be posed as a regularized inverse problem and solved by ensemble Kalman methods through the calibration of an eddy-diffusivity mass-flux scheme for subgrid-scale turbulence and convection, using data generated by large-eddy simulations. We find the algorithms amenable to batching strategies, robust to noise and model failures, and efficient in the calibration of hybrid parameterizations that can include empirical closures and neural networks.

show abstract

Section: Application To An Atmospheric Subgrid-scale Modelmentioning

confidence: 99%

“… Note. The prior mean values are taken from LG2020 (Lopez‐Gomez et al., 2020), C2020 (Cohen et al., 2020), and H2021 (He et al., 2021), where a physical description of the parameters may be found. …”

Section: Application To An Atmospheric Subgrid‐scale Modelmentioning

confidence: 99%

Training Physics‐Based Machine‐Learning Parameterizations With Gradient‐Free Ensemble Kalman Methods

Lopez‐Gomez

Christopoulos

Ervik

et al. 2022

J Adv Model Earth Syst

Self Cite

View full text Add to dashboard Cite

show abstract

“…This runtime was chosen to be much longer than the equilibration time of the SCM to the steady forcing; experiments using a runtime of 12 Table 1: Parameters φ considered for calibration in this study. The prior mean values are taken from LG2020 (Lopez-Gomez et al, 2020), C2020 (Cohen et al, 2020) and H2021 (He et al, 2021), where a physical description of the parameters may be found.…”

Section: Description Of Les Data and Model Configurationsmentioning

confidence: 99%

Training physics-based machine-learning parameterizations with gradient-free ensemble Kalman methods

Lopez‐Gomez

Christopoulos

Ervik

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Most machine learning applications in Earth system modeling currently rely on gradientbased supervised learning. This imposes stringent constraints on the nature of the data used for training (typically, residual time tendencies are needed), and it complicates learning about the interactions between machine-learned parameterizations and other components of an Earth system model. Approaching learning about process-based parameterizations as an inverse problem resolves many of these issues, since it allows parameterizations to be trained with partial observations or statistics that directly relate to quantities of interest in long-term climate projections. Here we demonstrate the effectiveness of Kalman inversion methods in treating learning about parameterizations as an inverse problem. We consider two different algorithms: unscented and ensemble Kalman inversion. Both methods involve highly parallelizable forward model evaluations, converge exponentially fast, and do not require gradient computations. In addition, unscented Kalman inversion provides a measure of parameter uncertainty. How learning about parameterizations can be posed as an inverse problem and solved by ensemble Kalman methods is illustrated through the calibration of an eddy-diffusivity mass-flux scheme for subgridscale turbulence and convection, using data generated by large-eddy simulations. We find the algorithms amenable to batching strategies, robust to noise and model failures, and efficient in the calibration of hybrid parameterizations that can include empirical closures and neural networks.

show abstract

“…The extended eddy diffusivity mass flux (extended‐EDMF) scheme (a close cousin of the multifluid model that extends traditional EDMF schemes to include transience; Tan et al, 2018) has seen substantial progress in recent years, with the formulation of new closures and successful single‐column simulations of various moist convective regimes, including shallow convection (Cohen et al, 2020; He et al, 2020; Lopez‐Gomez et al, 2020). Rapid progress has also been made in developing the multifluid approach, including new numerical methods (Weller and McIntyre, 2019), new closures (Weller et al, 2020), and accurate simulations of dry convection (Thuburn et al, 2019; Weller et al, 2020; Shipley et al, 2022).…”

Section: Introductionmentioning

confidence: 99%

A two‐fluid single‐column model of turbulent shallow convection. Part III: Results and parameter sensitivity

McIntyre

Efstathiou

Thuburn

2022

Quart J Royal Meteoro Soc

View full text Add to dashboard Cite

Two-fluid modelling has recently emerged as a promising approach to representing cumulus convection in weather and climate models. This study applies the two-fluid model described in Part II (Thuburn et al., 2022b) to a shallow cumulus convection case study over land (ARM). Large eddy simulation data is used to tune the majority of the closures that determine the properties of entrained and detrained air. The two-fluid model is generally able to reproduce the profiles of the mean and turbulent quantities over all stages of the diurnal cycle. As such, the initiation of shallow convection and the evolution of the cloud layer are well captured. The robustness of the two-fluid model is further verified using a steady-state test case (BOMEX), in which the cloud properties are also well modelled.

show abstract

An Improved Perturbation Pressure Closure for Eddy-Diffusivity Mass-Flux Schemes

Cited by 5 publications

References 50 publications

Training Physics‐Based Machine‐Learning Parameterizations With Gradient‐Free Ensemble Kalman Methods

Training Physics‐Based Machine‐Learning Parameterizations With Gradient‐Free Ensemble Kalman Methods

Training physics-based machine-learning parameterizations with gradient-free ensemble Kalman methods

A two‐fluid single‐column model of turbulent shallow convection. Part III: Results and parameter sensitivity

Contact Info

Product

Resources

About