A stochastic modeling framework to generate 2-D rough-wall high-Reynolds-number turbulent boundary layers

arXiv:2609.10236 · physics.flu-dyn · Submitted 2026-09-09 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: We're looking at the paper: "A stochastic modeling framework to generate 2-D rough-wall high-Reynolds-number turbulent boundary layers".

Mira: Here's the gist of the paper. A 2-D stochastic based model is developed for generating instantaneous velocity profiles featuring key elements of wall turbulent flows, such as uniform momentum zones and shear layers and vortices.

Kai: This framework is extended to reproduce the high-Reynolds-number, rough-wall, turbulent boundary layer measured by Saddoughi and Veeravalli without a UMZ dataset using only four critical flow parameters.

Mira: Model Architecture and Generation The stochastic generation of an organized 2-D rough-wall turbulent boundary layer velocity field involves a three-step procedure.

Paper summary: Kai: So looking at the paper titled "A stochastic modeling framework to generate two-D rough-wall high-Reynolds-number turbulent boundary layers", what we see is a systematic approach to building these complex flow structures stochastically <ref:2609.10236#pg1,A stochastic modeling framework to generate 2-D rough-wall high-Reynolds>. It claims this method can generate a near wall two dimensional velocity field that extends into the logarithmic layer and reproduces both the moments of the flow and the spectral characteristics of the streamwise velocity component using only four critical parameters.

Mira: The implication here for us is that if you have those four parameters, you can create a synthetic field for a rough-wall turbulent boundary layer without needing any existing dataset for uniform momentum zones or vortices to be present. This is useful because it lets researchers explore flow conditions where experimental data might be scarce, as long as they are willing to rely on the model's assumptions about those structures.

Lev: From an error correction perspective, the limitation we need to focus on is that this framework is two dimensional and not time dependent. So when you try to run it on real hardware or in a full simulation environment, you have to be careful about how much fidelity you expect for the three dimensional behavior and how much energy deficit you’re willing to accept from that spatial concatenation algorithm.

Kai: It seems like the work is really about proving that this process is scalable enough to generate a structured field that informs larger simulations, even if it's not perfectly capturing everything in three dimensions right away. The paper shows they can increase the spectral estimate of the dissipation rate from twenty-seven m squared per second cubed in their baseline rough wall high Reynolds number velocity field model up to fifty-six m squared per second cubed at a wall normal location of z over delta equal point zero nine for the HRVF plus vortex case.

Mira: That jump in the dissipation rate estimate shows that adding those near wall prograde vortices really helps improve the model's accuracy in that specific region. But they also pointed out a key limitation: you cannot know the required modeling parameters a priori, which means this is not a fully automated process for any arbitrary flow condition yet.

Lev: So what future work looks like is taking this foundation and trying to expand it from two dimensional to a fully three dimensional divergence free domain. That would involve characterizing and integrating the spanwise statistics of those uniform momentum zones. That’s where the real next step seems to be for moving this into a more comprehensive framework for large eddy simulations.

Kai: It’s about taking a structurally informed architecture that can initialize those simulations, even if it's still just a starting point, because building that initial field is often the hardest part of the simulation itself.

Conclusion: Kai: So this paper is about building a computer model that can create these messy, rough turbulent boundary layers using randomness instead of just looking at existing data.

Mira: Right, so they're taking some basic flow parameters—things like friction velocity and roughness length—and using a three-step process to generate a full two dimensional velocity field.

Lev: From my side, the real thing is that this whole process has to be run on a computer in real time or something, because we need to see how it handles those structures when you try to make it work on actual hardware.

Kai: The authors say they can generate a near wall two dimensional field that goes all the way into the logarithmic layer and even match the statistical moments of the flow.

Mira: But they did this without having any pre-existing data sets for things like uniform momentum zones or vortices, which is pretty ambitious for rough wall flows.

Lev: That's where I get cautious; if you want to run this on real hardware, you have to worry about how much fidelity you actually get when those synthetic structures are superimposed.

Kai: The main finding is that adding these near wall prograde vortices helps improve the model's estimate of the energy dissipation rate by boosting it from twenty seven m squared per second cubed in their baseline case up to fifty six m squared per second cubed at a wall normal distance of zero point zero nine delta.

Mira: That jump shows that those extra vortices are actually doing something important for how we measure turbulence in that near wall area.

Lev: But the caveat is they admit this whole framework is two dimensional and it doesn't evolve over time, which means you can't really test its ability to handle three dimensional behavior yet.

Kai: So the big picture here is that this architecture might be useful for initializing large eddy simulations because it gives you a starting point for a flow condition where data is scarce.

Mira: It’s about creating a way to generate plausible physics on demand, even if you can't perfectly calibrate every single parameter before you run the simulation.

Lev: And that leads into their next plan, which is trying to expand this from two dimensions into a fully three dimensional divergence free domain by figuring out how those uniform momentum zones behave in three dimensions.

Saint Anthony Falls Laboratory, University of Minnesota

physics.flu-dyn

Submitted: 2026-09-09

Updated: 2026-11-12

Comments: 29 pages, 7 figures

Code: https://github.com/Roozbeh96/SG-of-Velocity-field-Full-Implementation

License: http://creativecommons.org/licenses/by-nc-sa/4.0/

Importance score: 82/100

The gist: A 2-D stochastic based model is developed for generating instantaneous velocity profiles featuring key elements of wall turbulent flows, such as uniform momentum zones and shear layers and vortices.

Key concepts

Stochastic Generation of 1-D Step-Like Velocity Profiles (SGVP)
This step creates 1D velocity profiles by generating random numbers from a uniform distribution for each zone. This process starts at a specific wall elevation and sets the initial characteristics for the different flow regions, such as where momentum zones might exist.
Correlated 2-D Velocity Field (LRVF)
This technique takes the generated 1D profiles and arranges them into a 2D field. The arrangement is based on sorting these profiles according to how strongly their fluctuating streamwise velocity components are correlated, ensuring a realistic spatial structure.
Vortex Superimposition (SGVorX)
Synthetic sub-grid vortices are added onto the shear layers and in the near-wall region. This addition helps improve the flow statistics, specifically mitigating an underestimation of velocity variance observed very close to the wall.

Terminology

Summary

A 2-D stochastic based model is developed for generating instantaneous velocity profiles featuring key elements of wall turbulent flows, such as uniform momentum zones and shear layers and vortices. p. 0

Where the paper says it

We have developed a 2-D stochasticbased model for the generation and the streamwise concatenation of instantaneous, steplike velocity profiles featuring key elements of wall turbulent flows, i. , uniform momentum zones (UMZ) and shear layers [1, 2], and vortices [3].

This framework is extended to reproduce the high-Reynolds-number, rough-wall, turbulent boundary layer measured by Saddoughi and Veeravalli without a UMZ dataset using only four critical flow parameters. p. 0
Where the paper says it

, uniform momentum zones (UMZ) and shear layers [1, 2], and vortices [3]. The model is extended herein to the top of the logarithmic layer to reproduce the high-Reynolds-number, rough-wall, turbulent boundary layer measured by Saddoughi and Veeravalli [4], without the support of a UMZ dataset, using only a handful of critical flow parameters: the Taylor microscale λT , the boundary layer height δ, the friction velocity uτ , and the aerodynamic roughness length z0.

Model Architecture and Generation

The stochastic generation of an organized 2-D rough-wall turbulent boundary layer velocity field involves a three-step procedure.

  1. Step 1 involves Stochastic Generation of 1-D Step-Like Velocity Profiles (SGVP), which requires generating three random numbers from a uniform distribution for each zone's characteristics. The generation sequence is initiated at a prescribed wall-normal elevation, zstart = 1. p. 3

    Where the paper says it

    ARC Geophysical Research (2025) 1, — • For the step-like velocity profiles, the streamwise velocity component (u) is represented interchangeably by the zone’s modal velocity (um). 2 Algorithm The stochastic generation of an organized 2-D rough-wall turbulent boundary layer velocity field, within the logarithmic region of the wall, featured by vortices seeded within the shear layers and near-wall regions, is achieved via a three-step procedure: Step 1: Stochastic Generation of 1-D Step-Like Velocity Profiles (SGVP): To stochastically generate a step-like velocity profile, comprising multiple stacked UMZs, the generation sequence is initiated at a prescribed wall-normal elevation, zstart = 1.

    5ks.

  2. Step 2 constructs a correlated 2-D velocity field (LRVF) from the generated 1-D profiles by sorting them based on the maximum cross-correlation of the fluctuating streamwise velocity component. p. 3

    Where the paper says it

    The comprehensive wallnormal statistical analysis of the UMZ characteristics and the underlying profile-generation framework are detailed in Ehsani et al. Step 2: Stochastic Generation of a 2-D Velocity Field (SGVF): To construct a correlated 2-D velocity field, hereafter termed a low-resolution velocity field (LRVF), from the stochastically generated 1-D velocity profiles generated in Step 1, a sorting algorithm is implemented.

  3. Step 3 involves Stochastic Generation of a Vortex (SGVorX), where synthetic sub-grid vortices are superimposed onto the shear layers and within the immediate near-wall region. p. 5

    Where the paper says it

    Following the resolution enhancement, synthetic sub-grid vortices are superimposed onto the shear layers and within the immediate near-wall region. To construct the j-th synthetic vortex instance, parameterized by its core radius rωj , maximum azimuthal velocity uωj , and a structural transformation parameter ρ u,w ωj , three random numbers are drawn from a uniform distribution: {randrωj ,randuωj ,randρ u,w ωj } ∼ U(0, 1).

Parameterization and Scaling

The framework is governed by fundamental boundary layer parameters: friction velocity uτ, aerodynamic roughness length z0, boundary layer thickness δ, and Taylor microscale λT. These four primary parameters allow the suite of stochastic models to generate a near-wall 2-D velocity field extending into the logarithmic layer and reproduce both the moments of the flow and the spectral characteristics of the streamwise velocity component. p. 2

Where the paper says it

) of the UMZs, shear layers, and vortex attributes require the friction velocity uτ , boundary layer thickness δ, Taylor microscale λT , and aerodynamic roughness length z0 ≈ ks/30, where ks represents the equivalent sand-grain roughness. Therefore, given these four primary parameters for a target rough-wall turbulence flow field, this suite of stochastic models can generate a near-wall 2-D velocity field extending into the logarithmic layer and is able to satisfactorily reproduce both the moments of the flow and the spectral characteristics of the streamwise velocity component.

The Taylor microscale is estimated using statistics at z/δ = 0. p. 6

Where the paper says it

The reported friction velocity (uτ ), boundary layer thickness (δ), aerodynamic roughness length (z0), and Taylor microscale (λT ) serve as the primary scaling inputs to initialize the stochastic framework. The Taylor microscale is estimated using the statistics at z/δ = 0.

36. The storage size M is formulated as a baseline scaling relationship of M = 10√Reτ. p. 7
Where the paper says it

By optimizing this parameter across multiple datasets [10, 37], Ehsani et al. [2] formulated a baseline scaling relationship of M = 10√ Reτ .

Statistical Validation and Enhancements

The resulting turbulence statistics are compared against the experimental results, with the model successfully reproducing the logarithmic law of the wall except in the immediate near-wall region. The introduction of near-wall prograde vortices is shown to mitigate an underestimation of the normalized streamwise velocity variance u′2/u2τ observed below z/δ ≈ 0. p. 1

Where the paper says it

Identifying the correct distributions of UMZ attributes and employing UMZs’ statistical properties along the wall-normal direction (mean µ(z) and standard deviation σ(z)), extracted from experimental Particle Image Velocimetry (PIV) [37] and super-large-scale particle image velocimetry (SLPIV) [10] datasets, independent 1-D modal velocity profiles (step-like) were generated via the inverse transform sampling technique. The ensemble averages of the first- and secondorder statistical moments of the generated profiles were observed to reproduce the logarithmic law of the wall (except in the immediate near-wall region) and yield acceptable levels of velocity variance and covariance [1].

02. Furthermore, the application of a triangular band-stop filter is introduced to suppress residual excess energy observed at the wavenumber scale corresponding to k1 ∼ 2π/λ. p. 7
Where the paper says it

25 herein, the present framework introduces a dual-modification strategy: • Storage Capacity Expansion: The storage capacity M is enlarged to allow for increased spatial correlations between step-velocity profiles extending further from the wall and thus possessing a higher degree of freedom. • Spectral Notch Filtering: A triangular band-stop filter is applied directly to the streamwise and cross-stream velocity signals to partially suppress the residual excess energy observed at the wavenumber scale corresponding to k1 ∼ 2π/λ, as detailed in Section 3.

Performance and Limitations

The analysis shows that the vortex superimposition increases the spectral estimate of the dissipation rate, ϵ = 15νRk 21E11(k1)dk1, from 27[m2/s3] in the baseline HRVF model to 56[m2/s3] at a wall-normal location of z/δ = 0.09 for the HRVF+VorX case. p. 5

Where the paper says it

The remainder of this paper is organized as follows: § 2 introduces the model parameters; § 3 outlines modifications to the framework, including studies on storage size M, and the implementation of a band-stop filter; § 4 validates the resulting energy spectra and statistical moments against the benchmark dataset; and § 6 provides concluding remarks and future horizons for LES wall-modeling deployment. 2 Model Parameters To validate the model, the baseline parameters are chosen to match the high-speed boundary layer case documented by Saddoughi and Veeravalli [4]; these values are summarized in Table 1.

Despite these successes, limitations include the inability to know the required modeling parameters a priori, and the model being two-dimensional and not evolving in time. p. 4
Where the paper says it

The emerging mixed scaling between the viscous length scale (lν) and a wall-attached length scale (kz) suggests a weak dependency on the elevation that could be overlooked in a relatively shallow layer where the above approximations may hold. In view of a low-dimensional modeling approach, it can be argued that the Taylor microscale could be estimated by reducing the strictly required parameters to z0, uτ and δ.

The spatial concatenation algorithm induces an energy deficit in the vertical velocity component, affecting the cross-stream variance w′2/u2τ and the large-scale vertical velocity spectrum E22. p. 7
Where the paper says it

• Spectral Notch Filtering: A triangular band-stop filter is applied directly to the streamwise and cross-stream velocity signals to partially suppress the residual excess energy observed at the wavenumber scale corresponding to k1 ∼ 2π/λ, as detailed in Section 3. While the Reynolds number dependency of storage size is documented, the impact of expanding the vertical extent of the generated profiles on horizontal field coherence has not been 8

Conclusion and Future Work

The main outcome of the current work is that these processes are reasonably described and scalable, well enough that a synthetic field can be reconstructed for an untested flow condition without requiring a supporting UMZ or vortex calibration dataset. Future extensions of this framework will focus on expanding the stochastically generated velocity field from a two-dimensional to a fully three-dimensional divergence-free domain by characterizing and integrating the spanwise statistics of the uniform momentum zones. p. 1

Where the paper says it

[2] to construct a correlated two-dimensional (2-D) velocity field by grouping similar step-like velocity profiles together. Spectral analysis of the stochastically generated velocity field demonstrated the reproduction of turbulence structures ranging from the energy-containing eddies to the inertial subrange, capturing the k −5/3 1 regime.

The scalable, structurally informed architecture is envisioned to serve as an advanced wall-modeling framework for initializing large eddy simulations. p. 1
Where the paper says it

With recent advancements in numerical methods and computational power, large-eddy simulations (LES) have become an established tool for modeling the ASL and enabling predictive capabilities, thus providing a viable, affordable alternative to field experiments [6, 11–13]. While wall-resolved large-eddy simulation (WRLES) provides high fidelity [14], its computational cost is, however, notoriously prohibitive for high-Reynoldsnumber applications; the required number of grid points scales as ∼ Re13/7 τ [15, 16].

Code Availability

The underlying source code for the stochastic generation of the wall turbulent boundary layer flow and statistical extraction algorithms is openly available on GitHub at github.com/Roozbeh96/SG-of-Velocity-field-Full-Implementation. p. 6

Where the paper says it

Those functions, listed in Tables 2 and 3, were able to reasonably reproduce the wind tunnel and ASL UMZ statistics, but have never been used in the velocity profile generation process. The most relevant novelty of the present contribution is the implementation of the predictive models of height-dependent UMZ and vortex attributes for an untested flow condition without a supporting multi-dimensional UMZ and vortex datasets [4], and the validation of the resulting modal velocity field.

The repository includes scripts for generating step-like UMZ profiles, smoothing shear layers, reorganization to construct a correlated 2-D velocity field, resolution enhancement, vortex generation and insertion in the shear layers and near-wall region, and spectral and spatial analysis of the generated velocity field. p. 3
Where the paper says it

2 Algorithm The stochastic generation of an organized 2-D rough-wall turbulent boundary layer velocity field, within the logarithmic region of the wall, featured by vortices seeded within the shear layers and near-wall regions, is achieved via a three-step procedure: Step 1: Stochastic Generation of 1-D Step-Like Velocity Profiles (SGVP): To stochastically generate a step-like velocity profile, comprising multiple stacked UMZs, the generation sequence is initiated at a prescribed wall-normal elevation, zstart = 1. 5ks, identified at the interface between the roughness sublayer and the logarithmic region where UMZ attributes can be collected.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Stochastic modeling for generative turbulence generation: The improved system can generate a two-dimensional rough-wall turbulent boundary layer velocity field using only four critical flow parameters (Taylor microscale lambdaT, boundary layer height delta, friction velocity u tau, and aerodynamic roughness length z0).

  2. Prediction of near-wall turbulent statistics: The model can reproduce the high-Reynolds-number rough-wall turbulent boundary layer measured by Saddoughi and Veeravalli without the need for a Uniform Momentum Zone dataset.

  3. Validation against experimental data: The system can validate its generated statistical moments, energy spectra, and structure functions against experimental results using only the four critical flow parameters, as demonstrated by reproducing the high-Reynolds-number wind tunnel measurements of Saddoughi and Veeravalli.

  4. Resolution enhancement capabilities: The framework can increase spatial resolution from a low-resolution velocity field to a high-resolution velocity field by introducing synthetic subgrid vortices into shear layers and near-wall regions, refining the grid spacing to lambdaT over 10 times.

  5. Spectral energy distribution control: The system can modulate the energy distribution across scales by employing a triangular band-stop filter to suppress residual excess energy observed within the intermediate inertial subrange.

Abstract

Atmospheric surface layer flows are computationally challenging, predominantly due to surface roughness and the high Reynolds number, both of which demand exceptionally high spatial resolution in the near-surface region. We have developed a 2-D stochastic-based model for the generation and the streamwise concatenation of instantaneous, step-like velocity profiles featuring key elements of wall turbulent flows, i.e., uniform momentum zones (UMZ) and shear layers (Ehsani et al. 2024a,b), and vortices (Ehsani et al. 2026). The model is extended herein to the top of the logarithmic layer to reproduce the high-Reynolds-number, rough-wall, turbulent boundary layer measured by Saddoughi and Veeravalli(1994), without the support of a UMZ dataset, using only a handful of critical flow parameters: the Taylor microscale λT, the boundary layer height δ, the friction velocity uτ, and the aerodynamic roughness length z0. The primary challenge lies in the integration of the stochastic model with the scaled distributions of the UMZ and vortex characteristics. The resulting statistical moments, energy spectra, and structure functions are compared against the experimental results. The validated code is made available in a GitHub repository.

Related papers