Turbulence Closure With Small, Local Neural Networks: Forced Two‐Dimensional and β‐Plane Flows

Journal of Advances in Modeling Earth Systems, Journal Year: 2024, Volume and Issue: 16(4)

Published: April 1, 2024

Abstract We parameterize sub‐grid scale (SGS) fluxes in sinusoidally forced two‐dimensional turbulence on the β ‐plane at high Reynolds numbers (Re ∼25,000) using simple 2‐layer convolutional neural networks (CNN) having only O(1000) parameters, two orders of magnitude smaller than recent studies employing deeper CNNs with 8–10 layers; we obtain stable, accurate, and long‐term online or a posteriori solutions 16× downscaling factors. Our methodology significantly improves training efficiency speed large eddy simulations runs, while offering insights into physics closure such turbulent flows. approach benefits from extensive hyperparameter searching learning rate weight decay coefficient space, as well use cyclical annealing, which leads to more robust accurate compared fixed rates. either coarse velocity vorticity strain fields inputs, output components deviatoric stress tensor, S d . minimize loss between SGS flux divergence (computed high‐resolution solver) that obtained CNN‐modeled , without requiring energy enstrophy preserving constraints. The success shallow accurately parameterizing this class flows implies stresses have weak non‐local dependence fields; it also aligns our physical conception small‐scales are locally controlled by larger scales vortices their strained filaments. Furthermore, CNN‐parameterizations likely be interpretable.

Language: Английский

---

Generative Convective Parametrization of a Dry Atmospheric Boundary Layer

Journal of Advances in Modeling Earth Systems, Journal Year: 2024, Volume and Issue: 16(6)

Published: June 1, 2024

Abstract Turbulence parametrizations will remain a necessary building block in kilometer‐scale Earth system models. In convective boundary layers, where the mean vertical gradients of conserved properties such as potential temperature and moisture are approximately zero, standard ansatz which relates turbulent fluxes to via an eddy diffusivity has be extended by mass‐flux for typically asymmetric up‐ downdrafts atmospheric layer. We present parametrization dry transiently growing layer based on generative adversarial network. The training test data obtained from three‐dimensional high‐resolution direct numerical simulations. model incorporates physics self‐similar growth following classical mixed theory Deardorff renormalization. This enhances base machine learning algorithm thus significantly improves predicted statistics synthetically generated turbulence fields at different heights inside layer, above surface Differently stochastic parametrizations, our is able predict highly non‐Gaussian transient buoyancy fluctuations, velocity, flux also capturing fastest thermals penetrating into stabilized top region. results agree with two‐equation schemes. provides additionally granule‐type horizontal organization convection cannot any other closures. Our proof concept‐study paves way efficient data‐driven natural flows.

Language: Английский

---

Mesoscale Eddy‐Induced Sharpening of Oceanic Tracer Fronts

Journal of Advances in Modeling Earth Systems, Journal Year: 2025, Volume and Issue: 17(3)

Published: March 1, 2025

Abstract Oceanic fronts are ubiquitous and important features that form evolve due to multiscale oceanic atmospheric processes. Large‐scale temperature tracer fronts, such as those found along the eastward extensions of Gulf Stream Kuroshio currents, crucial components regional ocean environment climate. This numerical study examines relative importance large‐scale currents mesoscale (“eddies”) in front formation evolution. Using an idealized model double‐gyre system on both eddy‐resolving coarse‐resolution grids, we demonstrate effect eddies is sharpen front, whereas current counteracts this acts create a broader front. The eddy‐driven frontogenesis further described terms recently proposed framework generalized eddy‐induced advection, which represents all eddy effects tracers not mass fluxes traditionally parameterized by isopycnal diffusion. In advection formulated using effective velocity (EEIV), speed at move contours. advantage formulation frontal sharpening can be readily reproduced EEIVs. A functional EEIV variables effectively simulation. shows promise for advective parameterize models eddy‐resolving.

Language: Английский

---

Data-Driven Stochastic Closure Modeling via Conditional Diffusion Model and Neural Operator

Journal of Computational Physics, Journal Year: 2025, Volume and Issue: unknown, P. 114005 - 114005

Published: April 1, 2025

Language: Английский

---

Simulating Atmospheric Processes in Earth System Models and Quantifying Uncertainties With Deep Learning Multi‐Member and Stochastic Parameterizations

Journal of Advances in Modeling Earth Systems, Journal Year: 2025, Volume and Issue: 17(4)

Published: April 1, 2025

Abstract Deep learning is a powerful tool to represent subgrid processes in climate models, but many application cases have so far used idealized settings and deterministic approaches. Here, we develop stochastic parameterizations with calibrated uncertainty quantification learn convective turbulent surface radiative fluxes of superparameterization embedded an Earth System Model (ESM). We explore three methods construct parameterizations: (a) single Neural Network (DNN) Monte Carlo Dropout; (b) multi‐member parameterization; (c) Variational Encoder Decoder latent space perturbation. show that the parameterization improves representation processes, especially planetary boundary layer, compared individual DNNs. The respective illustrates are advantageous dropout‐based DNN regarding spread processes. Hybrid simulations our best‐performing remained challenging crash within first days. Therefore, pragmatic partial coupling strategy relying on for condensate emulation. Partial reduces computational efficiency hybrid Earth‐like enables model stability over 5 months parameterizations. However, exhibit biases thermodynamic fields differences precipitation patterns. Despite this, enable improvements reproducing tropical extreme traditional convection parameterization. these challenges, results indicate potential new generation machine leveraging improve stochasticity effects.

Language: Английский

---

Uncertainty Quantification of a Machine Learning Subgrid‐Scale Parameterization for Atmospheric Gravity Waves

Journal of Advances in Modeling Earth Systems, Journal Year: 2024, Volume and Issue: 16(7)

Published: July 1, 2024

Abstract Subgrid‐scale processes, such as atmospheric gravity waves (GWs), play a pivotal role in shaping the Earth's climate but cannot be explicitly resolved models due to limitations on resolution. Instead, subgrid‐scale parameterizations are used capture their effects. Recently, machine learning (ML) has emerged promising approach learn parameterizations. In this study, we explore uncertainties associated with ML parameterization for GWs. Focusing training process (parametric uncertainty), use an ensemble of neural networks emulate existing GW parameterization. We estimate both offline raw NN output and online model output, after coupled. find that parametric uncertainty contributes significant source must considered when introducing This quantification provides valuable insights into reliability robustness ML‐based parameterizations, thus advancing our understanding potential applications modeling.

Language: Английский

---

Subgrid Parameterizations of Ocean Mesoscale Eddies Based on Germano Decomposition

Journal of Advances in Modeling Earth Systems, Journal Year: 2023, Volume and Issue: 15(10)

Published: Oct. 1, 2023

Abstract Ocean models at intermediate resolution (1/4 ° ), which partially resolve mesoscale eddies, can be seen as Large eddy simulations of the primitive equations, in effect unresolved eddies must parameterized. In this work, we propose new subgrid that are consistent with physics two‐dimensional flows. We analyze fluxes barotropic decaying turbulence using Germano (1986, https://doi.org/10.1063/1.865568 ) decomposition. show Leonard and Cross stresses responsible for enstrophy dissipation, while Reynolds stress is additional kinetic energy (KE) backscatter. utilize these findings to a model, consisting three parts, compared baseline dynamic Smagorinsky model. The three‐component model accurately simulates spectral transfer improves representation KE spectrum, resolved decay posteriori experiments. backscattering component (Reynolds stress) implemented both quasi‐geostrophic equation ocean statistical characteristics, such vertical profile KE, meridional overturning circulation cascades potential energy.

Language: Английский

---

Updates on Model Hierarchies for Understanding and Simulating the Climate System: A Focus on Data‐Informed Methods and Climate Change Impacts

Journal of Advances in Modeling Earth Systems, Journal Year: 2023, Volume and Issue: 15(10)

Published: Oct. 1, 2023

Abstract The climate model hierarchy encompasses models of varying complexity along different axes, ranging from idealized that elegantly describe isolated mechanisms to fully coupled Earth system aspire provide useable projections. Based on the second Model Hierarchies Workshop, which took place in 2022, we present perspectives how this field has evolved since first Workshop 2016. In period, have witnessed a dramatic increase use (a) machine learning modeling and (b) estimate risks influence decision making under change. Here, discuss implications these growing areas research expect them become integrated into hierarchies framework.

Language: Английский

---

Learning Atmospheric Boundary Layer Turbulence

Authorea (Authorea), Journal Year: 2023, Volume and Issue: unknown

Published: June 23, 2023

Accurately representing vertical turbulent fluxes in the planetary boundary layer is vital for moisture and energy transport. Nonetheless, parameterization of remains a major source inaccuracy climate models. Recently, machine learning techniques have gained popularity oceanic atmospheric processes, yet their high dimensionality limits interpretability. This study introduces new neural network architecture employing non-linear reduction to predict dry convective layer. Our method utilizes kinetic scalar profiles as input extract physically constrained two-dimensional latent space, providing necessary minimal information accurate flux prediction.We obtained data by coarse-graining Large Eddy Simulations covering broad spectrum conditions, from weakly strongly unstable. These regimes are employed constrain space disentanglement, enhancing By applying this constraint, we decompose various scalars into two main modes variability: wind shear transport.Our data-driven accurately predicts (heat passive scalars) across regimes, surpassing state-of-the-art schemes like eddy-diffusivity mass scheme. projecting each variability mode onto its associated gradient, estimate diffusive learn eddy diffusivity. The found be significant only surface both becomes negligible mixed retrieved diffusivity considerably smaller than previous estimates used conventional parameterizations, highlighting predominant non-diffusive nature

Language: Английский

---

Stochastic Latent Transformer: Efficient Modeling of Stochastically Forced Zonal Jets

Journal of Advances in Modeling Earth Systems, Journal Year: 2024, Volume and Issue: 16(6)

Published: June 1, 2024

Abstract We present a novel probabilistic deep learning approach, the “stochastic latent transformer” (SLT), designed for efficient reduced‐order modeling of stochastic partial differential equations. Stochastically driven flow models are pertinent to diverse range natural phenomena, including jets on giant planets, ocean circulation, and variability midlatitude weather. However, much recent progress in has predominantly focused deterministic systems. The SLT comprises stochastically‐forced transformer paired with translation‐equivariant autoencoder, trained toward Continuous Ranked Probability Score. showcase its effectiveness by applying it well‐researched zonal jet system, where interaction between stochastically forced eddies mean results rich low‐frequency variability. accurately reproduces system dynamics across various integration periods, validated through quantitative diagnostics that include spectral properties rate transitions distinct states. achieves five‐order‐of‐magnitude speedup emulating zonally‐averaged compared direct numerical simulations. This acceleration facilitates cost‐effective generation large ensembles, enabling exploration statistical questions concerning probabilities spontaneous transition events.

Language: Английский

---