Pith. sign in

REVIEW 3 major objections 6 minor 72 references

Enhanced State Estimation for turbulent flows combining Ensemble Data Assimilation and Machine Learning

T0 review · 3 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read ML surrogates trained by data assimilation match EnKF accuracy for turbulent channel flow

desk verdict Useful DA-trained RFR surrogate for coarse IBM channel flow, but the pointwise state-estimation surrogate is not shown to represent the nonlocal Kalman update. read the letter →

arxiv 2501.18262 v1 pith:NFYISAL3 submitted 2025-01-30 physics.flu-dyn

Miguel M. Valero , Marcello Meldi This is my paper · ORCID
classification physics.flu-dyn
keywords EnKFRandomforestregressionTurbulentchannelflowImmersedboundarymethodDataassimilationStateestimationMachine-learningsurrogateReynoldsnumbertransferability
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proposes training Random Forest Regression models on the corrections produced by an ensemble Kalman filter (EnKF) during data-assimilation analysis phases, so that the cheap learned surrogates can be deployed inside a coarse immersed-boundary simulation after the sensor stream stops. In a turbulent plane channel flow at Re_tau ≈ 550, the ML-augmented simulation reproduces the mean velocity, friction velocity, and most Reynolds-stress components of the full data-assimilation run at a fraction of its computational cost. This matters because sequential assimilation degrades once observations are no longer available, whereas a trained surrogate can keep producing accurate state estimates over long time windows. The methodology transfers to finer near-wall grids and to Re_tau = 395, but the paper also reports failure at Re_tau = 950 when the grid no longer resolves the viscous sublayer.

What carries the argument

The central object is a decomposition of the EnKF analysis update u_a = u_f + K(Y − H(u_f)) into two learnable maps, each approximated by a Random Forest. The IBM RFR takes the local forecast velocity components and wall-normal distance, normalized by the friction velocity, and outputs the streamwise penalty forcing that the data-assimilation-optimized immersed boundary method would produce. The SE RFR takes the local forecast velocity vector and outputs the correction Δu applied by state estimation. Both maps are embedded in the PISO algorithm of the OpenFOAM solver, so each time step applies the learned penalty forcing and the learned state correction without computing ensemble covariances or a Kalman gain. The coupling tool CONES is used to stream analysis-phase data to the training pipeline.

What would settle it

Run the trained SE RFR surrogate inside a coarse IBM simulation of a turbulent channel flow at a Reynolds number where the wall-normal grid resolves the viscous sublayer, and compare its instantaneous correction field with the true EnKF analysis increment computed from the same forecast and observations; if the pointwise Random Forest correction cannot reproduce the non-local, covariance-weighted EnKF increment at each analysis time, the surrogate is only capturing a time-averaged bias and the central claim fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that the nontrivial corrections made by ensemble data assimilation can be distilled by machine learning into a cheap, deployable surrogate that reproduces the accuracy of the full assimilation without requiring live sensor data. Concretely, the EnKF's analysis phase produces two kinds of corrections—optimized immersed-boundary forcing coefficients and velocity-field updates—and Random Forest Regressors trained on those corrections, when run inside the same coarse solver, match the first- and second-order statistics of the full EnKF-augmented simulation for the turbulent plane channel flow. The key demonstration is that the ML surrogate remains accurate after the observation window ends, where a purely parameter-optimized model visibly degrades; this removes the main practical limitation of sequential assimilation. The same surrogate also performs well at a different wall-normal grid resolution and at a nearby Reynolds number, but it fails when the near-wall flow physics fall outside the training data.

Load-bearing premise

The central assumption is that the correction a flow cell needs at any future moment can be predicted from just that cell's current velocity and its distance from the wall, whereas the data-assimilation update it replaces uses information from the whole ensemble and from many sensors.

Editorial extensions

If this is right

  • A single trained ML surrogate can be embedded in a coarse immersed-boundary solver and reproduce ensemble-assimilation accuracy after the sensor stream has stopped, enabling flows observable only briefly—such as engine or turbomachinery passages—to be monitored over long windows.
  • The ML-augmented simulation runs at roughly four to five times lower elapsed time than the full data-assimilation run while matching its first-order statistics, making high-accuracy state estimation affordable for repeated or real-time use.
  • The same learned surrogates transfer to different grid resolutions when the near-wall resolution is at least as fine as in training, and they transfer to a nearby Reynolds number (Re_tau = 395) with a small friction-velocity error.
  • When the wall resolution is too coarse to resolve the viscous sublayer, as in the Re_tau = 950 test, the ML surrogate degrades; the paper explicitly concludes the training set must include the target y+ regime or a wall model.
  • The surrogate formulation is an additive correction to the forecast, so it can be inserted into an existing solver with only a minor modification to the momentum equation and pressure-velocity loop.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the surrogate is pointwise and the flow is statistically steady, the paper's success may depend on the EnKF correction being nearly stationary; unsteady flows with phase-locked coherent structures would likely require time-delay coordinates or a recurrent structure to capture evolving covariances.
  • The reported degradation of the velocity frequency spectra relative to the full data-assimilation run suggests the Random Forest may be reproducing a statistically averaged correction rather than the instantaneous covariance-weighted update; a direct comparison between the inferred correction field and the true EnKF increment at matching times would settle this.
  • The recipe of using data assimilation to generate labeled correction data, training a cheap surrogate, and deploying it when observation stops is not limited to fluids; any sequential estimator with an analysis phase could be distilled this way as long as the correction law is stationary over the training window.
  • The failure at Re_tau = 950 implies the surrogate has learned wall-resolved physics rather than a universal correction; enriching the training set with wall-modelled or higher-y+ samples, as the paper itself suggests, is a concrete next step.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes a hybrid DA-ML strategy for turbulent channel flow at Reτ≈550. An EnKF is first used to generate a large dataset of analysis-phase corrections for a coarse-grid IBM simulation. Two random-forest surrogates are then trained: an IBM RFR that predicts the streamwise penalization forcing as a function of local velocity and wall distance, and an SE RFR that predicts the EnKF state-estimation increment from the local forecast velocity. These surrogates are embedded in the PISO solver and tested in-range, on modified grids, and at Reτ=395 and 950. The authors report that the combined ML tool matches the DA solution for low-order statistics at a fraction of the cost, while acknowledging spectral degradation and a clear failure at Reτ=950.

Significance. If the central issue below is resolved, this is a valuable practical contribution. The paper provides genuine out-of-sample evidence at a different grid resolution and at Reτ=395, reports the Reτ=950 limitation honestly, quantifies an order-of-magnitude cost reduction (C.C*≈8.8 for ML s.e. versus 37.6 for DA s.e.), and releases the CONES coupling tool, which aids reproducibility. These strengths are substantial. The main uncertainty is whether the SE RFR surrogate learns a local approximation to a fundamentally nonlocal Kalman update or merely reproduces a time-averaged correction profile, and the manuscript's own spectral evidence currently leans toward the latter interpretation.

major comments (3)
  1. [Section 4.3, Appendix B Eq. (B.5), Appendix A Eqs. (A.7)-(A.10)] The SE RFR surrogate defined in Eq. (B.5) is a strictly local map: the correction at a cell depends only on the three forecast velocity components at that cell and on the normalization. The EnKF analysis it replaces, Eqs. (A.7)-(A.10), is nonlocal: the Kalman gain K_k = L ⊙ X(S)^T [S S^T + R]^{-1} couples every grid cell through the ensemble anomaly matrices X and S and through the localization matrix L. No argument or experiment in Section 4.3 establishes that a pointwise conditional-mean map can represent this nonlocal operator for the present flow. The stationarity statement in Section 4.3 concerns time dependence, not spatial locality. The manuscript's own spectral evidence supports the concern: Section 5 states that the s.e. spectra are omitted because they show no significant deviations from the p.o. curves, and Fig. 10 shows the p.o. spectra to be in line with the unaugmented IBM/BF runs rather than with the DA reference. Please provide quantitative evidence on the locality question, for example by training the same RFR on a nonlocal feature set or by analyzing the spatial support of the Kalman increments for this configuration, or substantially soften the claim that SE RFR reproduces the DA state update.
  2. [Section 4.2, Fig. 10, Section 5, abstract] The headline claim that the ML tool matches the accuracy of the complete DA strategy is not supported for instantaneous dynamics. Fig. 10 shows DNS-IBM-MLp.o. spectra close to the DNS-IBM and DNS-BF runs and clearly below DNS-IBM-DAp.o.; Section 5 reports that the s.e. spectra were omitted because they do not deviate significantly from the p.o. curves. Thus the ML-augmented runs match the low-order statistics (mean velocity, Reynolds stresses, uτ) but lose the spectral accuracy of DA. This distinction should be made explicit in the abstract and conclusions, and the claimed equivalence should be restricted to the statistical moments actually validated.
  3. [Introduction vs. Section 4.3] The introduction claims that the global ML tool is not affected by the time window of investigation and by the availability of further data. This overstates the result: both RFR models are trained on the last 1.8 tA of the DA observation window (Section 4.3), after the EnKF has converged, and no training is possible without an observation window. Section 3.2 shows that DA parametric optimization alone degrades after the observation window ends, so the ML version cannot be trained from such a degraded window. The claim should be rephrased to refer to deployment after training for statistically stationary conditions, and the sensitivity to the length and placement of the training window should be stated.
minor comments (6)
  1. [Section 4.3] The definition of the third component of Y is incorrect: it reads (u^f_z,k − u^f_z,k)/uτ_k, which is identically zero; it should be (u^a_z,k − u^f_z,k)/uτ_k.
  2. [Section 4.3] The text refers to 'the black-box IBM RFR derived in §4.3'; the correct reference is Section 4.1.
  3. [Figure 10] The legend includes 'DNS-IBM-CF', a label that is never defined and does not appear in Table 1; please correct it or add a definition.
  4. [Section 5] The statement '5 at the top wall and 5 and the bottom wall' should read '5 at the top wall and 5 at the bottom wall.'
  5. [Figures 4, 8, 12, 14-17] The comparisons between ML, DA, and reference runs are presented only visually, without error bars or ensemble spreads. Reporting quantitative discrepancies with confidence intervals would make the claimed equivalence between ML and DA much easier to assess.
  6. [Tables 2 and 3] The R² and NRMSE diagnostics are computed on the training dataset or a held-out fraction of the same converged DA window; they do not test generalization. The out-of-grid and Reτ=395 tests are the real generalization evidence, and reporting quantitative errors for those cases would strengthen the paper.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the ML models are trained on DA outputs, but the central generalization claims are tested on independent grids and external DNS at Re_tau=395 and 950, so the predictions do not reduce by construction to their training targets.

full rationale

The derivation chain is explicit: DA produces training data (fP and Delta u = u_a - u_f from the EnKF analysis, Eqs. A.7-A.10), RFR models are trained on those targets (Secs. 4.1, 4.3, Eqs. B.2, B.5), and the trained surrogates are coupled with the low-fidelity IBM solver. In-training-condition agreement (Sec. 4.2: 'The black-box IBM RFR is now assessed using conditions consistent with the training data') is a surrogate fit-quality check, not a physical prediction. The load-bearing predictive claims are tested out of sample in Sec. 5: the same trained RFR is applied to different grids (IBM-ML-grid 1/2) and to Re_tau = 395 and 950, against Moser et al. and Hoyas et al. DNS; ML-ReTau395 matches the external reference and ML-ReTau950 fails, so the comparison is falsifiable. Self-citations to Valero & Meldi [48] and CONES [12] supply the DA baseline and coupling code, but they are not invoked as uniqueness theorems or to define the ML output, and the prior DA results were validated against external DNS in [48]. The pointwise SE RFR (Eq. B.5) versus nonlocal Kalman gain (Eq. A.7) is an unproven representational assumption, i.e. a correctness risk, and is consistent with the paper's acknowledged omissions/failures ('spectral analysis is omitted' in Sec. 4.3; Re_tau = 950 degradation in Sec. 5), but it is not circularity: the model is not defined in terms of the quantity it claims to predict. No step reduces by construction to its inputs.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The method's free parameters are mostly carried over from the authors' prior DA work [48] or set by hand; the central claims depend on these values. The most consequential assumption is the unexamined locality of the state-estimation surrogate.

free parameters (7)
  • IBM penalization coefficient D (initial value) = D = 10^5
    Uniform initial penalization coefficient following Valero & Meldi [48]; the DA optimizes its local values, and the RFR surrogate is trained on the optimized forcing.
  • Observation error covariance R = not specified
    Gaussian observation uncertainty is assumed, but the numerical values of R are not reported in the paper, although the EnKF analysis depends on them.
  • Localization length scales eta_x, eta_y, eta_z = not specified
    Tuning factors in Eq. A.8 controlling the spatial extent of the Kalman gain; no values are given in the manuscript.
  • Covariance inflation factor lambda = lambda = 1.01 (for Section 5 runs)
    Multiplicative covariance inflation used to counteract filter underdispersion (Appendix A). The value for the original DA run is not stated.
  • Ensemble size Ne = Ne = 40 (for Section 5 runs)
    Number of ensemble members used in the DA analysis and, indirectly, in the RFR training; it controls the correction statistics the surrogate learns.
  • RFR hyperparameters = R = 100 trees, min leaf size = 5, feature fraction = 1/3
    Hyperparameters selected for computational efficiency; the paper states preliminary tests showed weak sensitivity, but no sensitivity analysis is reported.
  • Training window for IBM RFR = 76 snapshots over 1.5 tA (m_T = 6,225,920 samples)
    Training data is taken from the final 1.5 tA of the DA observation window; the choice of window length is ad hoc.
assumptions (5)
  • standard math Incompressible Navier-Stokes equations for Newtonian fluids (Eqs. 1-2)
    Governing equations for all simulations in the paper.
  • domain assumption Volume penalization IBM with a scalar tensor D = D I (Eq. 3)
    Penalization scheme used to enforce the no-slip condition at immersed walls; the optimized coefficients are the target of the DA and ML training.
  • domain assumption EnKF modeling assumptions: Gaussian, unbiased, independent observation errors; ensemble represents the true error covariance (Appendix A)
    Standard EnKF assumptions inherited from [1]; the paper does not validate them for this flow.
  • domain assumption Statistical stationarity of the flow: the action of the DA tool is statistically the same over the observation window (Section 4.3)
    Justifies training the state-estimation surrogate on 15 analysis phases and then applying it beyond the observation window; valid only for statistically steady flows.
  • ad hoc to paper Pointwise locality of the state-estimation correction: SE RFR uses only local velocity components and wall distance as inputs (Appendix B, Eq. B.5)
    Assumes the EnKF update, which is non-local through the Kalman gain and localization (Eq. A.7), can be represented as a cell-local map. This is not derived or analyzed in the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhanced State Estimation for turbulent flows combining Ensemble Data Assimilation and Machine Learning." pith.science (2026). https://pith.science/paper/NFYISAL3

@misc{pith2026250118262,
  author       = {Pith},
  title        = {Pith review of: Enhanced State Estimation for turbulent flows combining Ensemble Data Assimilation and Machine Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NFYISAL3}},
  note         = {Machine review of arXiv:2501.18262}
}
abstract

A novel strategy is proposed to improve the accuracy of state estimation and reconstruction from low-fidelity models and sparse data from sensors. This strategy combines ensemble Data Assimilation (DA) and Machine Learning (ML) tools, exploiting their complementary features. ML techniques rely on the data produced by DA methods during analysis phases to train physics-informed corrective algorithms, which are then coupled with the low-fidelity models when data from sensors is unavailable. The methodology is validated via the analysis of the turbulent plane channel flow test case for $Re_\tau \approx 550$. Here, the low-fidelity model consists of coarse-grained simulations coupled with the Immersed Boundary Method (IBM), while observation is sampled by a highly refined body-fitted calculation. The analysis demonstrates the capabilities of the algorithm based on DA and ML to accurately predict the flow features with significantly reduced computational costs. This approach exhibits potential for future synergistic applications of DA and ML, leveraging the robustness and efficiency of ML models alongside the physical interpretability ensured by DA algorithms.

Figures

Figures reproduced from arXiv: 2501.18262 by the authors.

Figure 1
Figure 1. Difference between the state augmentation and parametric optimisation via [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Schematic representation of a parent node and the criteria to split it into two [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. (a) Scheme of the turbulent plane channel flow, and (b) distribution of the cell layers in the wall-normal direction for the solid region Ωb, the fluid region Ωf , and the interface regions Σb regions for the IBM simulations. element starting from the outer limit of the computational domain, where a no-slip boundary condition is imposed. As shown in [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Comparison of the main statistical moments of the velocity field. Results are [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]
Figure 5
Figure 5. Figure 5: Penalty coefficients Dij improved by means of ( ) DNS-IBM-DAs.e., concern￾ing the ones employed in the classical penalisation ( ) DNS-IBM. The standard deviation σij of the prior for the initial Gaussian perturbation of the ensemble is represented in light grey (σij = …
Figure 6
Figure 6. Figure 6: Averaged penalty source term ⟨f P ⟩ of the database employed for training. In light grey, we represent the standard deviation. Test sample i input (u ∗ xi , u∗ yi , u∗ zi , y∗ i ) Tree 1 Tree 2 (· · ·) Tree r Tree R . . . . . . . . . ˆf ∗ Pxi , 1 ˆf ∗ Pxi , 2 (· · ·) ˆ…
Figure 7
Figure 7. Figure 7: Overview of RFR algorithm for the prediction of the streamwise forcing term [PITH_FULL_IMAGE:figures/full_fig_p020_7.png]
Figure 8
Figure 8. Figure 8: Comparison of the main statistical moments of the velocity field. Results are [PITH_FULL_IMAGE:figures/full_fig_p021_8.png]
Figure 9
Figure 9. Figure 9: Isocontours of Q-criterion calculated for [PITH_FULL_IMAGE:figures/full_fig_p023_9.png]
Figure 10
Figure 10. Figure 10: F F T calculated by sampling the fluctuating velocity field u ′ located at (first row) ∆y ⋆ ≈ 30 and (second row) ∆y ⋆ ≈ 56. Results are shown for the simulations ( ) DNS-IBM-MLp.o., ( ) DNS-IBM-DAp.o., ( ) DNS-IBM-CF, ( ) DNS-BF, and ( ) R-DNS-BF. long time window. H…
Figure 11
Figure 11. Figure 11: Overview of the DA code CONES employed for state estimation and parameter [PITH_FULL_IMAGE:figures/full_fig_p026_11.png]
Figure 12
Figure 12. Figure 12: Comparison of the main statistical moments of the velocity field. Results [PITH_FULL_IMAGE:figures/full_fig_p027_12.png]
Figure 13
Figure 13. Figure 13: Visualisation of the near-wall streaks for [PITH_FULL_IMAGE:figures/full_fig_p028_13.png]
Figure 14
Figure 14. Figure 14: Comparison of the main statistical moments of the velocity field. Results are [PITH_FULL_IMAGE:figures/full_fig_p031_14.png]
Figure 15
Figure 15. Figure 15: Comparison of the main statistical moments of the velocity field. Results are [PITH_FULL_IMAGE:figures/full_fig_p032_15.png]
Figure 16
Figure 16. Figure 16: Comparison of the main statistical moments of the velocity field for [PITH_FULL_IMAGE:figures/full_fig_p034_16.png]
Figure 17
Figure 17. Figure 17: Comparison of the main statistical moments of the velocity field. Results are [PITH_FULL_IMAGE:figures/full_fig_p035_17.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

72 extracted references · 67 canonical work pages

  1. [1]

    M. Asch, M. Bocquet, M. Nodet, Data Assimilation: Methods and Ap- plications, SIAM, 2016

    work page 2016

  2. [2]

    Evensen, F

    G. Evensen, F. Vossepoel, P. J. van Leeuwen, Data Assimilation Fun- damentals, Springer Textbooks in Earth Sciences, Geography and En- vironment, 2022

    work page 2022

  3. [3]

    Burkov, The Hundred-Page Machine Learning Book, Andriy Burkov, 2019

    A. Burkov, The Hundred-Page Machine Learning Book, Andriy Burkov, 2019

    work page 2019

  4. [4]

    Brunton, B

    S. Brunton, B. Noack, P. Koumoutsakos, Machine learning for fluid mechanics, Annual Review for Fluid Mechanics 52 (2020) 477–508

    work page 2020

  5. [5]

    Gorl´ e, G

    C. Gorl´ e, G. Iaccarino, A framework for epistemic uncertainty of turbu- lent scalar flux models for Reynolds-averaged Navier-Stokes simulations, Physics of Fluids 25 (2013) 055105

    work page 2013

  6. [6]

    W. N. Edeling, P. Cinnella, R. P. Dwight, H. Bijl, Bayesian estimates of parameter variability in the K–ε turbulence model, Journal of Com- putational Physics 258 (2014) 73–94

    work page 2014

  7. [7]

    Margheri, M

    L. Margheri, M. Meldi, M. V. Salvetti, P. Sagaut, Epistemic uncertain- ties in RANS model free coefficients, Computers & Fluids 102 (2014) 315–335. 40

    work page 2014

  8. [8]

    Tracey, K

    B. Tracey, K. Duraisamy, J. Alonso, A paradigm for data-driven predic- tive modeling using field inversion and machine learning, 53rd AIAA Aerospace Sciences Meeting Session: Turbulence Modeling I (2015) 2015–1287

    work page 2015

Show all 72 references
  1. [9]

    J. L. Wu, H. Xiao, E. Paterson, Physics-informed machine learning ap- proach for augmenting turbulence models: A comprehensive framework, Physical Review Fluids 7 (2018) 074602

    2018

  2. [10]

    Srinivasan, L

    P. Srinivasan, L. Guastoni, H. Azizpour, P. Schlatter, R. Vinuesa, Pre- dictions of turbulent shear flows using deep neural networks, Physical Review Fluids 4 (2019) 054603

    2019

  3. [11]

    P. S. Volpiani, M. Meyer, L. Franceschini, J. Dandois, F. Renac, E. Mar- tin, O. Marquet, D. Sipp, Machine learning-augmented turbulence mod- eling for rans simulations of massively separated flows, Phys. Rev. Fluids 6 (2021) 064607. doi:10.1103/PhysRevFluids.6.064607

    2021 doi

  4. [12]

    Villanueva, M

    L. Villanueva, M. M. Valero, A. S. Glumac, M. Meldi, Augmented state estimation of urban settings using on-the-fly sequential data assimila- tion, Comp. Fluids 269 (2024) 106118

    2024

  5. [13]

    Meldi, D

    M. Meldi, D. Lucor, P. Sagaut, Is the Smagorinsky coefficient sensitive to uncertainty in the form of the energy spectrum?, Physics of Fluids 23 (2011) 125109

    2011

  6. [14]

    Vollant, G

    A. Vollant, G. Balarac, C. Corre, Subgrid-scale scalar flux modelling based on optimal estimation theory and machine-learning procedures, Journal of Turbulence 18 (9) (2017) 854–878

    2017

  7. [15]

    M. Meldi, Augmented Prediction of Turbulent Flows via Sequential Es- timators: Sensitivity of State Estimation to Density of Time Sampling for Available Observation, Flow, Turbulence and Combustion 101 (2018) 389–412

    2018

  8. [16]

    Chandramouli, E

    P. Chandramouli, E. Memin, D. Heitz, 4D large scale variational data assimilation of a turbulent flow with a dynamics error model, Journal of Computational Physics 412 (2020) 109446

    2020

  9. [17]

    Lozano-Dur´ an, J

    A. Lozano-Dur´ an, J. Bae, Self-critical machine-learning wall-modeled les for external aerodynamics, Annual Research Briefs 2020 - (2020) 1–15. 41

    2020

  10. [18]

    V. Mons, Y. Du, T. Zaki, Ensemble-variational assimilation of statistical data in large-eddy simulation, Physical Review Fluids 6 (2021) 104607

    2021

  11. [19]

    Moldovan, A

    G. Moldovan, A. Mariotti, G. Lehnasch, L. Cordier, M. Salvetti, M. Meldi, Multigrid sequential data assimilation for the large-eddy sim- ulation of a massively separated bluff-body flow, Computers and Fluids 281 (2024) 106385

    2024

  12. [20]

    Villanueva, K

    L. Villanueva, K. Truffin, M. Meldi, Synchronization and optimization of large eddy simulation using an online ensemble kalman filter, International Journal of Heat and Fluid Flow 110 (2024) 109597. doi:10.1016/j.ijheatfluidflow.2024.109597. URL https://linkinghub.elsevier.com...

    2024

  13. [21]

    Plogmann, O

    J. Plogmann, O. Brenner, P. Jenny, Adjoint-based assimilation of sparse time-averaged velocity data in large-eddy simulations (2024). arXiv: eprint={2407.11746}

    2024

  14. [22]

    Molinaro, J.-S

    R. Molinaro, J.-S. Singh, S. Catsoulis, C. Narayanan, D. Lakehal, Em- bedding data analytics and CFD into the digital twin concept, Comp. Fluids 214 (2021) 104759

    2021

  15. [23]

    D. Wagg, K. Worden, R. Barthorpe, P. Gardner, Digital twins: State-of- the-art and future directions for modelling and simulation in engineering dynamics applications, Amer. Soc. Mech. Eng. 6 (3) (2020) 030901

    2020

  16. [24]

    B. Yang, S. Yang, Z. Lv, F. Wang, T. Olofsson, Application of digital twins and metaverse in the field of fluid machinery pumps and fans: A review, Sensors 22 (2022) 9294

    2022

  17. [25]

    Sharma, E

    A. Sharma, E. Kohasih, J. Zhang, A. Brintrup, A. Calinescu, Digital twins: State of the art theory and practice, challenges, and open research questions, J. Ind. Inf. Int. 30 (2022) 100383

    2022

  18. [26]

    Schena, P

    L. Schena, P. Marques, R. Poletti, S. Ahizi, J. V. der Berghe, M. Mendez, Reinforcement twinning: From digital twins to model-based reinforce- ment learning, Journal of Computational Physics 82 (2024) 102421. 42

    2024

  19. [27]

    S. S. H. Boosari, Predicting the dynamic parameters of multiphase flow in cfd (dam-break simulation) using artificial intelligence-(cascading de- ployment), Fluids 4 (2019) 44

    2019

  20. [28]

    White, D

    C. White, D. Ushizima, C. Farhat, Fast neural network predictions from constrained aerodynamics datasets, AIAA Scitech 2020 Forum (2020)

    2020

  21. [29]

    Kochkov, J

    D. Kochkov, J. Smith, A. Alieva, Q. Wang, M. Brenner, S. Hoyer, Machine learning–accelerated computational fluid dynamics, Applied Mathematics 118 (21) (2021)

    2021

  22. [30]

    X. Fan, D. Akhare, J.-X. Wang, Neural differentiable modeling with diffusion-based super-resolution for two-dimensional spatiotemporal tur- bulence (2024). arXiv:2406.20047

    2024 arXiv

  23. [31]

    Evensen, The ensemble kalman filter for combined state and param- eter estimation, IEEE Control Systems Magazine (2009) 83–104

    G. Evensen, The ensemble kalman filter for combined state and param- eter estimation, IEEE Control Systems Magazine (2009) 83–104

    2009

  24. [32]

    Brajard, A

    J. Brajard, A. Carrassi, M. Bocquet, L. Bertino, Combining data assim- ilation and machine learning to infer unresolved scale parametrization, Phil. Trans. R. Soc. A 379 (2021) 20200086

    2021

  25. [33]

    Farchi, P

    A. Farchi, P. Laloyaux, M. Bonavita, M. Bocquet, Using machine learn- ing to correct model error in data assimilation and forecast applications, Quaterly Journal of the Royal Meteorological Society 147 (2021) 3067– 3084

    2021

  26. [34]

    Arcucci, J

    R. Arcucci, J. Zhu, S. Hu, Y.-K. Guo, Deep data assimilation: Integrat- ing deep learning with data assimilation, Applied Sciences 11 (1114) (2021)

    2021

  27. [35]

    Villiers, V

    R. Villiers, V. Mons, D. Sipp, E. Lamballais, M. Meldi, Enhancing unsteady reynolds-averaged navier-stokes modelling from sparse data through sequential data assimilation and machine learning, Flow, Tur- bulence and Combustion (2024)

    2024

  28. [36]

    Z. Y. Wang, W. W. Zhang, A unified method of data assimilation and turbulence modeling for separated flows at high reynolds numbers (2022). arXiv:2211.00601. 43

    2022 arXiv

  29. [37]

    Quattromini, M

    M. Quattromini, M. A. Bucci, S. Cherubini, O. Semeraro, Active learn- ing of data-assimilation closures using graph neural networks (2023). arXiv:2303.03806

    2023 arXiv

  30. [38]

    Y. Ling, A. Lozano-Dur´ an, Numerically consistent data-driven subgrid- scale model via data assimilation and machine learning, AIAA SciTech Forum (2025)

    2025

  31. [39]

    Breiman, Random forests, Machine Learning 45 (2001) 5–32

    L. Breiman, Random forests, Machine Learning 45 (2001) 5–32

    2001

  32. [40]

    Angot, C.-H

    P. Angot, C.-H. Bruneau, P. Fabric, A penalization method to take into account obstacles in incompressible viscous flows, Numer. Math. 81 (1999) 497–520

    1999

  33. [41]

    Verzicco, Immersed boundary methods: Historical perspective and future outlook, Annual Review of Fluid Mechanics 55 (2022) 129 – 155

    R. Verzicco, Immersed boundary methods: Historical perspective and future outlook, Annual Review of Fluid Mechanics 55 (2022) 129 – 155

    2022

  34. [42]

    J. Kou, A. H. de Mendoza, S. Joshi, S. L. Clainche, E. Ferrer, Eigenso- lution analysis of immersed boundary method based on volume penal- ization: applications to high-order schemes, Journal of Computational Physics 449 (2022) 10817

    2022

  35. [43]

    Ferziger, M

    J. Ferziger, M. Peric, Computational Methods in Fluid Dynamics, New- York : Springer-Verlag, 1996

    1996

  36. [44]

    Greenshields, H

    C. Greenshields, H. Weller, Notes of Computational Fluid Dynamics: General Principles, CFD Direct Ltd, 2022

    2022

  37. [45]

    Issa, Solution of the implicitly discretised fluid flow equations by operator-splitting, Journal of Computational Physics 62 (1) (1986) 40–

    R. Issa, Solution of the implicitly discretised fluid flow equations by operator-splitting, Journal of Computational Physics 62 (1) (1986) 40–

    1986

  38. [46]

    H. K. Versteeg, W. Malalasekera, An Introduction to Computational Fluid Dynamics: The Finite Volume Method. Second Edition, Pearson Prentice Hall, 2007

    2007

  39. [47]

    openfoam.com

    OpenFOAM - Field Operation And Manipulation, https://www. openfoam.com

  40. [48]

    M. M. Valero, M. Meldi, An immersed boundary method using online sequential data assimilation, Journal of Computational Physics 524 (3 2025). doi:10.1016/j.jcp.2024.113697. 44

    2025

  41. [49]

    D. C. Wilcox, Turbulence Modeling for CFD, 3rd Edition, DCW Indus- tries Inc., La Canada CA, 2006

    2006

  42. [50]

    H. Xiao, P. Cinnella, Quantification of Model Uncertainty in RANS Simulations: A Review, Progress in Aerospace Sciences 108 (2019) 1– 31

    2019

  43. [51]

    Roßbach, Neural networks vs

    P. Roßbach, Neural networks vs. random forests - does it always have to be deep learning?, frankfurt School Blog (2018)

    2018

  44. [52]

    S. Wang, C. Aggarwal, H. Liu, Using a random forest to inspire a neu- ral network and improving on it, in: Proceedings of the 2017 SIAM International Conference on Data Mining, 2017

    2017

  45. [53]

    de Zordo-Banliat, G

    M. de Zordo-Banliat, G. Dergham, X. Merle, P. Cinnella, Space- dependent turbulence model aggregation using machine learning, Jour- nal of Computational Physics 497 (2024) 112628

    2024

  46. [54]

    Cherroud, X

    S. Cherroud, X. Merle, P. Cinnella, X. Gloerfelt, Space-dependent ag- gregation of data-driven turbulence models (2023). arXiv:2306.16996

    2023 arXiv

  47. [55]

    M. L. Kaandorp, R. P. Dwight, Data-driven modelling of the reynolds stress tensor using random forests with invariance, Comp. Fluid. 202 (2020) 104497

    2020

  48. [56]

    Wang, J.-L

    J.-X. Wang, J.-L. Wu, H. Xiao, Physics-informed machine learning ap- proach for reconstructing reynolds stress modeling discrepancies based on dns data, Physical Review Fluids 2 (2017) 034603

    2017

  49. [57]

    Heyse, A

    J. Heyse, A. Mishra, G. Iaccarino, Data driven physics constrained perturbations for turbulence model uncertainty estimation, in: AAAI Spring Symposium: MLPS, 2021

    2021

  50. [58]

    J. Kim, P. Moin, R. Moser, Turbulence statistics in fully developed channel flow at low reynolds number, Journal of Fluid Mechanics 177 (1987) 133–166

    1987

  51. [59]

    J. C. del Alamo, J. Jim´ enez, Spectra of the very large anisotropic scales in turbulent channels, Phys. Fluids 15 (6) (2003)

    2003

  52. [60]

    Hoyas, J

    S. Hoyas, J. Jim´ enez, Reynolds number effects on the reynolds-stress budgets in turbulent channels, Phys. Fluids 20 (2008) 101511. 45

    2008

  53. [61]

    Bernardini, S

    M. Bernardini, S. Pirozzoli, P. Orlandi, Velocity statistics in turbulent channel flow up to reτ = 4000, Journal of Fluid Mechanics 742 (2014) 171–191

    2014

  54. [62]

    Cimarelli, E

    A. Cimarelli, E. D. Angelis, P. Schlatter, G. Brethouwer, A. Talamelli, C. Casciola, Sources and fluxes of scale energy in the overlap layer of wall turbulence, Journal of Fluid Mechanics 771 (2015) 407–423

    2015

  55. [63]

    D. E. King, Dlib-ml: A machine learning toolkit, Journal of Machine Learning Research 10 (2009) 1755–1758

    2009

  56. [64]

    W. H. Press, S. A. Teukolsky, W. T. Vetterling, B. P. Flannery, Numeri- cal recipes: The Art of Scientific Computing. Third Edition, Cambridge University Press, 2017

    2017

  57. [65]

    doi:https://doi.org/10.1016/0021-9991(86)90099-9

    doi

  58. [66]

    G. I. Taylor, The spectrum of turbulence, in: Proc. R. Soc. Lond. A, Vol. 164, 1938, pp. 476–490

    1938

  59. [67]

    R. D. Moser, J. Kim, N. N. Mansour, Direct numerical simulation of turbulent channel flow up to reτ = 590, Physics of Fluids 11 (4) (1999) 943–945

    1999

  60. [68]

    Maejima, K

    S. Maejima, K. Tanino, S. Kawai, Physics-informed machine-learning solution to log-layer mismatch in wall-modeled large-eddy simulation, Physical Review Fluids 9 (2024) 084609

    2024

  61. [69]

    X. Yang, S. Zafar, J.-X. Wang, H. Xiao, Predictive large-eddy-simulation wall modeling via physics-informed neural networks, Physical Review Fluids 4 (2019) 034602

    2019

  62. [70]

    R. E. Kalman, A new approach to linear filtering and prediction prob- lems, Journal Basic Eng. 82 (1960) 35–45

    1960

  63. [71]

    Carrassi, M

    A. Carrassi, M. Bocquet, L. Bertino, G. Evensen, Data assimilation in the geosciences: An overview of methods, issues, and perspectives, WIREs Climate Change 9 (2018)

    2018

  64. [72]

    Hoteit, D.-T

    I. Hoteit, D.-T. Pham, M. E. Gharamti, X. Luo, Mitigating observation perturbation sampling errors in the stochastic enkf, Monthly Weather Review 143 (2015) 2918–2936. 46

    2015

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.