TL;DR
This paper introduces computationally efficient "abridged" statistical closures for simulating complex atmospheric and oceanic flows. These new methods, including the EDMIC model, closely match the accuracy of more computationally expensive methods and direct numerical simulations, making them promising for real-world applications.
What Problem Does This Paper Address?
Simulating the dynamics of turbulent flows in the atmosphere and oceans is crucial for weather forecasting and climate modeling. These flows are inherently complex, involving interactions between large-scale mean flows, smaller-scale turbulent eddies, Rossby waves (which are planetary-scale waves in rotating fluids), and topographical features like mountains.
Traditional methods for modeling these interactions, known as "closure theories," often lead to computationally intensive equations. Specifically, "non-Markovian" closures, which account for the memory of past states, require calculating extensive time-history integrals. This becomes a significant bottleneck for long-term simulations. The challenge is to develop closure theories that are both accurate and computationally efficient enough for practical use, especially for "inhomogeneous" flows (where properties vary in space) and when dealing with complex interactions like those involving Rossby waves and topography.
Methods
The authors build upon existing statistical dynamical closure theories, particularly the Quasi-diagonal Direct Interaction Approximation (QDIA). Their core innovation is the development of "abridged" versions of these closures. The key idea is to simplify the time-history integrals by replacing the complete trajectory of the mean flow over time with just its current-time value. This simplification is justified by the assumption that the mean field evolves more slowly than the turbulent fluctuations.
From this abridged QDIA, they formulate three variants of "Markovian Inhomogeneous Closures" (MICs). These MICs use different versions of the fluctuation-dissipation theorem (FDT) to approximate the information contained in the time-history integrals. Crucially, these abridged MICs require only two auxiliary "relaxation functions" to be computed, significantly reducing computational cost compared to previous methods that might need three.
A further advancement is the formulation of an "Eddy Damped Markovian Inhomogeneous Closure" (EDMIC). This model is a generalization of the well-established EDQNM closure for homogeneous turbulence. The EDMIC features analytical expressions for its relaxation functions, eliminating the need for auxiliary prognostic equations and making it even more computationally efficient.
Datasets
The authors used large ensembles of Direct Numerical Simulations (DNS) as a benchmark for evaluating their closure models. Specifically, they performed simulations with 1800 members to generate robust statistics for comparison. The experimental setup involved a large-scale eastward flow impinging on a conical mountain on a generalized beta-plane, designed to generate rapidly evolving Rossby waves and turbulence.
Key Insights and Ideas
- Simplification through "Abridging": Replacing the full time-history of the mean flow with its current value in closure equations dramatically reduces computational cost without sacrificing significant accuracy. This is a powerful simplification for simulating complex, evolving flows.
- MICs as Efficient Alternatives: The developed Markovian Inhomogeneous Closures (MICs) offer a computationally efficient way to approximate the behavior of non-Markovian closures, making long-term simulations more feasible.
- EDMIC: The Best of Both Worlds: The EDMIC model, with its analytical relaxation functions, represents a significant step towards highly efficient and accurate closures for inhomogeneous turbulence. It generalizes the successful EDQNM to more complex scenarios.
- Robustness to Simplification: The abridged closures, even with the simplification of the mean field, perform remarkably well, showing high pattern correlations with DNS. This suggests that the rapid evolution of the flow field leads to rapid decorrelation of perturbations, making the simplified approach effective.
- Sensitivity to FDT: The QDIA closure appears to be more sensitive to the approximation of the mean field than the MICs that also incorporate the current-time FDT.
Key Results
- The abridged QDIA and the three abridged MIC models showed excellent agreement with large ensembles of DNS for both mean flow fields and transient statistics.
- Pattern correlations between the abridged closures and DNS for the mean non-zonal streamfunction were very high, ranging from 0.9789 (QDIA) to 0.9999 (MIC with X=0).
- The EDMIC model, with its analytical relaxation functions, was formulated and shown to be realizable under similar conditions as the homogeneous EDQNM.
- The abridged closures were tested in simulations of a rapidly evolving mean flow interacting with topography and Rossby waves, demonstrating their ability to capture complex dynamics.
- The EDMIC model is suggested to be a natural generalization of the EDQNM for inhomogeneous turbulence, offering significant computational efficiency.
Limitations and Future Work
- Realizability: While the abridged MICs and EDMIC perform well, the authors note that in the presence of waves, their realizability (ensuring that physical quantities like energy remain non-negative) cannot always be guaranteed. They plan to address this in future work by generalizing the representations of time-dependent relaxation functions.
- Parameterization of Eddy Damping: For the EDMIC model, further work is planned to study the dependence on eddy damping parameters and to explore different parameterizations, especially for the D'(t) term, which is influenced by the mean field and topography.
- Extension to 3D: The authors suggest that the EDMIC closure could be extended to three-dimensional inhomogeneous turbulent flows with further effort in generalizing eddy damping parameterizations.
- Wave Renormalization: Future work will also aim to include wave renormalization effects to ensure realizability in the presence of time-dependent waves for both EDMIC and EDQNM models.
Summary updated