Markovian inhomogeneous closures for Rossby waves and turbulence over topography

Jorgen S. Frederiksen, T. Okane

Abstract

This paper establishes a family of manifestly Markovian inhomogeneous closures (MIC) designed to simulate the interaction of two-dimensional turbulent flows with Rossby waves and complex topography. Derived from the quasi-diagonal direct interaction approximation (QDIA), these closures address the computational bottlenecks associated with non-Markovian memory integrals. By modifying the response function and employing three distinct variants of the fluctuation-dissipation theorem (FDT)—current-time, prior-time, and correlation FDTs—the authors derive Markovian equations for triad relaxation functions. The performance of these closures is validated against large ensembles of direct numerical simulations (DNS) on a generalized beta-plane. Results demonstrate that the MIC models achieve exceptional accuracy, with pattern correlations exceeding 0.9998, while providing the computational efficiency required for long-term geophysical simulations.

Introduction

In geophysical fluid dynamics, modeling the interaction between turbulent flows and complex topography is essential for understanding atmospheric and oceanic circulation. Traditionally, researchers have relied on Direct Numerical Simulations (DNS) for high-fidelity results. However, the computational cost of DNS makes it impractical for the long integrations required in climate modeling. Statistical closures offer a more efficient alternative, but they often struggle with the inhomogeneous (non-uniform) nature of real-world flows, such as wind currents interacting with mountain ranges.

The gold standard for statistical accuracy in turbulence modeling has long been the Direct Interaction Approximation (DIA). While powerful, the DIA is non-Markovian, meaning it possesses a "memory" that requires integrating over the entire history of the flow to calculate its current state. This "time-history integral" creates a massive computational bottleneck. This research introduces Markovian Inhomogeneous Closures (MIC), which retain the structural sophistication of the DIA framework while eliminating the memory bottleneck, making them suitable for complex, large-scale simulations.

Theoretical Framework: From QDIA to MIC

The foundation of this work is the Quasi-Diagonal Direct Interaction Approximation (QDIA), a theory specifically developed to handle inhomogeneous flows. To achieve a Markovian form, the authors implement two critical modifications:

  1. Response Function Markovianization: The complex, history-dependent response function of the QDIA is replaced with a Markovian form that depends only on the current state.
  2. FDT Variants: The researchers explored three versions of the Fluctuation-Dissipation Theorem to relate internal fluctuations to dissipation:
  3. Current-time (quasi-stationary) FDT: Assumes the system is in a state of local equilibrium.
  4. Prior-time (non-stationary) FDT: Accounts for the evolving nature of the flow.
  5. Correlation FDT: Designed to ensure "realizability," preventing the model from producing unphysical results such as negative energy spectra.

By deriving differential equations for triad relaxation functions, the MIC replaces slow time-history integrals with efficient "tendency" equations. This allows the model to capture the essential physics of wave-turbulence interactions without the prohibitive storage and calculation costs of the full history.

Experimental Validation

To evaluate the MIC models, the authors compared them against a "ground truth" dataset consisting of an ensemble of 1,800 DNS runs. The test scenario involved a westerly mean flow impinging on a conical mountain on a generalized $\beta$-plane. This setup is a standard benchmark in geophysical fluid dynamics because it rapidly generates high-amplitude Rossby wavetrains—large-scale planetary waves—within a turbulent environment.

The simulations were conducted at a circular C16 resolution, which is sufficient to capture the non-linear interactions between the mean flow, the topography, and the transient eddies over a 10-day period. This "far-from-equilibrium" process provides a rigorous test for any statistical closure.

Key Results and Discussion

The performance of the MIC models was remarkably consistent with the DNS ensemble. Key findings include:

  • Exceptional Correlation: The pattern correlations for the mean Rossby wavetrains between the MIC models and the DNS ensemble were greater than 0.9998. This indicates that the Markovian approximation loses virtually no accuracy in capturing the spatial structure of the waves.
  • Spectral Fidelity: The evolved kinetic energy spectra—both for the mean flow and transient eddies—showed excellent agreement across all scales. The models accurately captured the transfer of energy between different wavenumbers as the Rossby waves amplified.
  • Realizability and Robustness: The version of the MIC utilizing the Correlation FDT proved particularly robust, ensuring that energy spectra remained non-negative even in the presence of high-amplitude waves.
  • Computational Efficiency: By bypassing the time-history integrals, the MIC models are significantly more efficient than the standard QDIA for long integrations, offering a path toward incorporating sophisticated closure physics into climate-scale models.

Conclusion and Future Directions

The development of Markovian Inhomogeneous Closures represents a significant step forward in the statistical modeling of geophysical turbulence. By successfully "Markovianizing" the QDIA, the authors have demonstrated that it is possible to maintain high levels of physical accuracy while drastically reducing computational overhead.

Future work will focus on extending these closures to higher Reynolds numbers, which may require empirical "regularization" to ensure small-scale power laws remain physically consistent. Additionally, the authors aim to develop analytical expressions for the triad relaxation functions, which would further enhance the efficiency of the MIC, potentially matching the speed of simpler isotropic closures like the Eddy-Damped Quasi-Normal Markovian (EDQNM) model while maintaining the ability to handle complex topography.

Summary updated

Abstract

Manifestly Markovian closures for the interaction of two-dimensional inhomogeneous turbulent flows with Rossby waves and topography are formulated and compared with large ensembles of direct numerical simulations (DNS) on a generalized beta plane. Three versions of the Markovian inhomogeneous closure (MIC) are established from the quasi-diagonal direct interaction approximation (QDIA) theory by modifying the response function to a Markovian form and employing respectively the current-time (quasi-stationary) fluctuation dissipation theorem (FDT), the prior-time (non-stationary) FDT and the correlation FDT. Markov equations for the triad relaxation functions are derived that carry similar information to the time-history integrals of the non-Markovian QDIA closure but become relatively more efficient for long integrations. Far from equilibrium processes are studied, where the impact of a westerly mean flow on a conical mountain generates large amplitude Rossby waves in a turbulent environment, over a period of 10 days. Excellent agreement between the evolved mean streamfunction and mean and transient kinetic energy spectra are found for the three versions of the MIC and two variants of the non-Markovian QDIA compared with an ensemble of 1800 DNS. In all cases mean Rossby wavetrain pattern correlations between the closures and the DNS ensemble are greater than 0.9998.
BibTeX
@inproceedings{frederiksen2019markovian,
  author = {Jorgen S. Frederiksen and T. Okane},
  title = {{Markovian inhomogeneous closures for Rossby waves and turbulence over topography}},
  booktitle = {EGU},
  year = {2019},
}
Figure