Semi-analytical eddy-viscosity and backscattering closures for 2D geophysical turbulence

This paper presents a novel semi-analytical derivation of parameters for Leith, Smagorinsky, and Jansen-Held closures in 2D geophysical turbulence, which, when applied to large-eddy simulations, accurately reproduce key DNS statistics and extreme events while outperforming existing dynamic and standard baseline models.

The Gist

Imagine you are trying to predict the weather or model the ocean currents. The Earth is a giant, swirling ball of fluid (air and water) that is incredibly complex. To understand it, scientists use supercomputers to simulate the movement of these fluids.

However, there's a problem: the Earth is huge, but computers are finite. We can't simulate every single tiny swirl of wind or ripple in the ocean because there are too many of them. It would take more computing power than exists in the universe.

So, scientists use a trick called Large Eddy Simulation (LES). Think of it like looking at a forest from a helicopter. You can see the big trees and the overall shape of the forest (the "large eddies"), but you can't see the individual leaves or the tiny bugs crawling on the bark (the "small eddies").

The Problem: The Missing Leaves

In our weather models, we simulate the big trees but ignore the leaves. But here's the catch: the tiny leaves (small-scale turbulence) actually affect the big trees. They steal energy from the big swirls and sometimes even push energy back into them.

If we just ignore the leaves, our model gets the energy balance wrong, and the weather forecast fails. We need a "closure"—a mathematical rule that guesses what the leaves are doing based on what we can see.

For decades, scientists have used "best guesses" (empirical parameters) for these rules. They'd say, "Let's try a number here, see if it works, and if not, try another." It's like tuning a radio by blindly turning the dial until the static clears up. It works, but it's messy and not very scientific.

The Breakthrough: A Semi-Analytical Recipe

This paper, by Yifei Guan and Pedram Hassanzadeh, is like finding the exact recipe for that radio dial instead of guessing.

They didn't just guess the numbers; they derived them mathematically using the fundamental laws of physics. They looked at how energy moves in 2D turbulence (like weather systems) and used a known pattern of how energy spreads out (called a "scaling law") to calculate the perfect numbers for their models.

Here is how they did it, using a simple analogy:

1. The "Energy Highway" (The Scaling Law)

Imagine energy in the atmosphere is like cars on a highway. In 2D turbulence, there's a specific rule for how fast cars move and how they spread out. The authors used this rule (a $k^{-3}$ law) to figure out exactly how much "friction" or "push" the small, invisible leaves should exert on the big trees.

2. The Three Models (The Tools)

They focused on three specific tools scientists use to fix the missing leaves:

• The Smagorinsky Model: A classic tool that acts like a sponge, soaking up extra energy to stop the simulation from exploding.
• The Leith Model: A more sophisticated sponge specifically designed for 2D flows (like weather).
• The Jansen-Held (JH) Model: The most advanced tool. It's a sponge that can also squeeze energy back out. Sometimes, tiny swirls push energy up to the big swirls (this is called "backscattering"). The JH model accounts for this "re-injection."

3. The "Magic Number" (Parameter A)

To get the perfect settings for these tools, the authors needed one piece of data: a constant called $A$.

• The Old Way: You'd have to run a massive, perfect simulation (DNS) for every single new weather scenario to find this number.
• The New Way: The authors discovered that for almost all 2D weather scenarios, this number $A$ is nearly the same (around 1.9). It's like realizing that almost all cars on the highway drive at roughly the same speed, regardless of the traffic.

So, instead of guessing or running expensive simulations, you can now use this "magic number" to calculate the perfect settings for your weather model instantly.

The Result: Why It Matters

The authors tested their new, mathematically derived numbers against:

1. Perfect simulations (the "truth").
2. Old methods (guessing the numbers).
3. Machine learning methods (using AI to find the best numbers).

The verdict? Their new math-based numbers were almost identical to the ones found by the expensive AI learning method.

What does this mean for you?

• Better Forecasts: Weather and climate models can now use these "perfectly tuned" settings without needing to run expensive AI training first.
• Extreme Events: These models are better at predicting rare, extreme events (like hurricanes or heatwaves) because they handle the energy transfers more accurately.
• Simplicity: It turns a complex, trial-and-error process into a straightforward calculation.

In a Nutshell

This paper is like taking a complex, messy recipe for a cake that required a chef to taste and adjust the sugar every time, and turning it into a precise, scientific formula. Now, anyone can bake the perfect cake (simulate the perfect weather) just by following the math, knowing that the ingredients (the parameters) are almost always the same. It makes our digital models of the Earth more accurate, faster, and more reliable.

Technical Summary

1. Problem Statement

Large-Eddy Simulation (LES) is essential for modeling geophysical turbulence (e.g., weather and climate prediction) where Direct Numerical Simulation (DNS) is computationally prohibitive. However, LES requires Subgrid-Scale (SGS) closures to model the effects of unresolved scales.

• Current Limitations: Existing closures, such as the Smagorinsky (3D), Leith (2D), and Jansen-Held (JH) models, rely on free parameters ($C_S$, $C_L$, $C_{JH}$, $C_B$) that are typically chosen empirically or via trial-and-error.
• The Gap: While analytical derivations exist for the Smagorinsky constant in 3D turbulence (linked to the Kolmogorov constant), no analytical derivations exist for the parameters of 2D turbulence closures ($C_L$, $C_{JH}$, $C_B$).
• Specific Challenge: 2D geophysical flows exhibit unique dynamics, including enstrophy cascades and significant "backscattering" (energy transfer from subgrid to resolved scales), which standard eddy-viscosity models fail to capture without specific backscattering terms.

2. Methodology

The authors employ a semi-analytical derivation approach to determine the closure parameters for 2D turbulence.

• Theoretical Framework:
  • The derivation assumes a high-Reynolds-number 2D turbulent flow.
  • It utilizes the turbulent kinetic energy (TKE) direct-cascade scaling law: $\hat{E}(k) = A \eta^{2/3} k^{-3}$, where $k$ is the wavenumber, $\eta$ is the enstrophy dissipation rate, and $A$ is a flow-dependent constant.
  • The authors assume that domain-averaged energy and enstrophy can be approximated by integrating this scaling law in the Fourier spectral domain up to the LES cutoff wavenumber ($k_c = \pi/\Delta$).
• Derivation Process:
  1. Leith Model ($C_L$): Derived by balancing the enstrophy transfer equation with the eddy-viscosity definition, linking the filter width $\Delta$ to the enstrophy-dissipation length scale.
  2. Jansen-Held Model ($C_{JH}$ and $C_B$): The JH model includes a biharmonic eddy-viscosity term (enstrophy dissipation) and an anti-diffusion term (backscattering). The authors derive $C_{JH}$ by balancing enstrophy transfer and derive $C_B$ by analyzing the ratio of backscattered energy to dissipated energy.
  3. Smagorinsky Model ($C_S$): Since a direct 2D analytical derivation for $C_S$ is lacking, the authors derive a relationship between $C_S$ and $C_L$ by equating the eddy viscosities of the Smagorinsky and Leith models for the same flow field.
  4. Extensions: The derivation is extended to include logarithmic corrections to the $k^{-3}$ spectrum (Kraichnan 1971) and general power-law spectra ($k^{-p}$).
• Validation:
  • The parameter $A$ is diagnosed from DNS data of 8 distinct 2D geophysical turbulence setups (varying Reynolds numbers, forcing wavenumbers, and $\beta$-effects).
  • The semi-analytically derived parameters are compared against Ensemble Kalman Inversion (EKI) optimized parameters (obtained via online learning in a previous study by the same authors) and standard baseline closures.

3. Key Contributions

• First Semi-Analytical Derivation: The paper provides the first semi-analytical formulas for the parameters $C_L$, $C_{JH}$, and $C_B$ in 2D geophysical turbulence.
• Unified Parameter Estimation: The study demonstrates that the constant $A$ in the TKE spectrum is nearly flow-independent (approximately 1.8–1.9) across diverse geophysical setups, except in cases with strong $\beta$-effects (strong anisotropy). This allows for a universal estimation of closure parameters without extensive trial-and-error.
• Backscattering Quantification: The derivation explicitly quantifies the backscattering parameter $C_B$ and its relationship to the dissipation parameter $C_{JH}$, showing how backscattering increases the required value of $C_{JH}$.
• Validation of Online Learning: The results validate the previous EKI-based findings, showing that the "optimal" parameters found via data-driven learning are consistent with physical scaling laws.

4. Results

• Parameter Agreement: The semi-analytically derived parameters ($C_L^{Analytical}$, $C_S^{Analytical}$, $C_{JH}^{Analytical}$) closely match the EKI-optimized parameters ($C_L^{EKI}$, etc.) across all 8 test cases.
  • $C_L$ and $C_S$ match within one standard deviation.
  • $C_{JH}$ matches within two standard deviations.
• Value of Constant A: The diagnosed value of $A \approx 1.8-1.9$ aligns well with previous renormalization group analysis ($A \approx 1.923$).
• LES Performance:
  • LES simulations using these semi-analytically derived parameters successfully reproduce key DNS statistics, including extreme events (tails of probability density functions) and interscale energy/enstrophy transfers.
  • These models significantly outperform baseline models (standard Smagorinsky/Leith and dynamic variants) in both a-priori (offline) and a-posteriori (online) tests.
• Specific Findings on $C_S$: The paper establishes a specific relationship between $C_S$ and $C_L$ for 2D turbulence, showing that $C_S$ depends weakly on the LES resolution due to logarithmic corrections, unlike the constant $C_S$ in 3D.

5. Significance

• Reduced Empiricism: This work moves SGS modeling for 2D geophysical turbulence away from purely empirical tuning toward a physics-based, semi-analytical framework.
• Computational Efficiency: By providing a method to estimate parameters from a few DNS snapshots (or even analytical renormalization group values), it eliminates the need for expensive online learning (EKI) for every new simulation setup.
• Improved Prediction Accuracy: The derived closures better capture the physics of backscattering and extreme events, which are critical for accurate weather and climate prediction.
• Theoretical Bridge: It bridges the gap between spectral scaling laws and practical LES implementation, offering a robust theoretical foundation for the Jansen-Held closure, which is increasingly used in global climate models.

In conclusion, Guan and Hassanzadeh successfully demonstrate that the parameters for 2D turbulence closures can be derived semi-analytically using fundamental spectral scaling laws, yielding results that are physically consistent and computationally superior to traditional empirical approaches.