Predominant-Mode Inversion of Surface Waves: Inherently Addressing Inconspicuous Low Frequency Mode Jumps

arXiv:2512.11265 · physics.geo-ph · Submitted 2025-12-12 · 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: I'm Kai.

Mira: And I'm Mira, here with Lev.

Kai: We're looking at the paper: "Predominant-Mode Inversion of Surface Waves".

Mira: Here's the gist of the paper. Predominant-mode inversion offers a robust and practical approach for surface-wave inversion at sites with strong impedance contrasts.

Kai: The predominant-mode framework inherently addresses low-frequency mode jumps by automatically identifying, at each frequency, the Rayleigh-wave mode with the maximum vertical surface amplitude.

Mira: This method allows inversions to proceed without the need for explicit mode indexing by the inversion analyst or unvalidated assumptions about fundamental modal dominance.

Kai: Addressing Modal Ambiguity The study introduces a new strategy where, instead of extracting only phase velocity, complete mode shapes are computed from the forward model to examine their amplitude distribution at the free surface.

Paper summary: Kai: So, we’re diving into the details of this paper, "Predominant-Mode Inversion of Surface Waves: Inherently Addressing Inconspicuous Low Frequency Mode Jumps". To start, the authors explain the core issue they are solving: inversion of Rayleigh wave dispersion data is really tough at sites with strong impedance contrasts where modal energy transitions smoothly from one mode to another at low frequencies.

Mira: They point out that analysts often misinterpret that smooth transition as just a continuation of the fundamental mode, which causes them to overestimate shear wave velocity in deeper layers or get the bedrock depth wrong.

Lev: That’s a common pitfall because when higher modes arise due to strong impedance contrast, they tend to be excited at lower frequencies, making them hard to detect in the experimental dispersion data.

Kai: The paper introduces this predominant-mode inversion framework as a way around that by automatically selecting the Rayleigh wave mode with the maximum vertical surface amplitude for every frequency you test.

Mira: They claim that this selection ensures an incorrect mode will inevitably lead to higher misfit because its phase velocity simply won't match what you recorded in the dispersion behavior.

Lev: So, instead of making a guess upfront about which mode is active, the method lets the data tell it which one is most likely present at each point in frequency space.

Kai: The methodology for this involves using the Thin Layer Method to turn the Rayleigh wave dispersion problem into an eigenvalue problem so they can calculate both the eigenvalues, which are phase velocities, and eigenvectors, which are mode shapes simultaneously.

Mira: The theoretical surface amplitude associated with a specific mode is written as a function of wavenumber, angular frequency, and the mode index.

Lev: And then the predominant mode is defined simply as the one with the largest vertical amplitude at the surface for each frequency that you are analyzing.

Kai: This definition is what makes it robust because it doesn't depend on any prior knowledge about which mode is dominant or how many modes there are.

Mira: It’s a clever way to handle modal ambiguity, and they show this method works even when the modal energy transition is smooth and continuous due to a lack of spatial resolution in the data.

Lev: And they also mention that by focusing on removing poorly constrained transition-zone frequencies, you stabilize the inversion process because it prevents conflicting modal behavior from happening during fitting.

Kai: The paper goes on to show how this framework handles the target dispersion data well across the usable frequency range, even when most of the inversion error is concentrated at a few isolated frequencies.

Mira: So what we’re seeing here is that you can get good velocity structure recovery even if there are some localized errors, which is better than getting a globally biased result where everything is just off.

Lev: It's about stabilizing the inversion by avoiding those transition frequency bands where the data gets messy.

Kai: This sets up nicely for us to talk about what this actually means for application and how it compares to other methods.

Conclusion: Kai: So we’ve covered a lot of ground on this paper, "Predominant-Mode Inversion of Surface Waves: Inherently Addressing Inconspicuous Low Frequency Mode Jumps". The authors are Mrinal Bhaumik and Brady R. Cox from Utah State University.

Mira: They essentially argue that the main problem with traditional methods is the difficulty in identifying which mode is active at low frequencies when there’s a smooth modal transition due to strong impedance contrasts.

Lev: And they propose this predominant-mode framework as a way to avoid making assumptions about mode dominance upfront and instead lets the data dictate what's dominant at each frequency.

Kai: The implication is that for someone listening, it means you can expect more stable velocity profile recovery when dealing with complex subsurface structures.

Mira: It means that if you’re working on a challenging site, this method provides a robust tool because it minimizes the risk of getting systematically biased results from mode misidentification.

Lev: And since they made the MATLAB code public, it means we can actually try to implement this approach ourselves and see how it runs in practice on real hardware.

Kai: So in short, this paper suggests that by focusing on finding the predominant mode at every frequency is a smart way to get more reliable velocity structure estimates.

Mrinal Bhaumik, Brady R. Cox

Utah State University

physics.geo-ph

Submitted: 2025-12-12

Updated: 2025-12-12

Journal ref: Engineering Geology, Vol. 369, July 2026, 108834

DOI: 10.1016/j.enggeo.2026.108834

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 83/100

The gist: Predominant-mode inversion offers a robust and practical approach for surface-wave inversion at sites with strong impedance contrasts.

Key concepts

Predominant Mode
Instead of choosing a specific wave type manually, this method automatically picks the Rayleigh wave mode that has the largest vertical surface amplitude at every measured frequency. This eliminates errors caused by incorrectly assuming which mode is dominant, leading to more reliable results.
Forward Modeling (Thin Layer Method)
This mathematical technique is used to simulate how surface waves behave based on subsurface properties. It treats the Rayleigh wave dispersion problem as an eigenvalue problem, allowing researchers to calculate both the phase velocities and the actual shape of the wave modes simultaneously.
Mode Jump
This refers to sudden changes in how surface waves behave at different frequencies, often indicating a change in which mode is dominant. The predominant-mode framework is specifically designed to handle these jumps without needing prior knowledge of where they occur.
Dispersion Misfit (md)
This metric measures how poorly the modeled dispersion curve matches the actual observed experimental data. The inversion seeks to minimize this value, using a weighted average of phase velocity differences between the model and the observations to find the best subsurface parameters.

Terminology

Summary

Predominant-mode inversion offers a robust and practical approach for surface-wave inversion at sites with strong impedance contrasts. p. 0 The predominant-mode framework inherently addresses low-frequency mode jumps by automatically identifying, at each frequency, the Rayleigh-wave mode with the maximum vertical surface amplitude. p. 0 This method allows inversions to proceed without the need for explicit mode indexing by the inversion analyst or unvalidated assumptions about fundamental modal dominance. p. 0

Addressing Modal Ambiguity

The study introduces a new strategy where, instead of extracting only phase velocity, complete mode shapes are computed from the forward model to examine their amplitude distribution at the free surface. The Rayleigh wave mode with the maximum vertical amplitude at the surface for each frequency is selected as the basis for inversion, which is referred to as the predominant mode. p. 3 This selection ensures that an incorrectly assigned mode will inevitably lead to higher misfit because its phase velocity does not match the recorded dispersion behavior and cannot adequately fit the experimental data. p. 3 Unlike the effective mode approach, which requires explicit knowledge of source offset, receiver spacing, and array geometry, the predominant mode does not depend on acquisition layout. p. 3

Forward Modeling Methodology

The predominant-mode forward model is derived using the Thin Layer Method (TLM). The TLM formulates the Rayleigh wave dispersion problem as an eigenvalue problem, allowing simultaneous computation of both the eigenvalues (phase velocities) and eigenvectors (mode shapes). p. 3 The theoretical surface amplitude associated with that mode at frequency is written as a function of wavenumber, angular frequency, and the mode index. p. 5 The predominant mode is defined as the one with the largest vertical amplitude at the surface for each frequency. p. 5

Performance on Synthetic Models

The predominant-mode method was integrated into a particle swarm optimization global search algorithm and used to invert three synthetic models exhibiting low-frequency mode jump behavior. Across all cases, the predominant-mode method accurately recovered major velocity contrasts and interface depths, whereas fundamental-mode inversions consistently overestimated Vs and mislocated deeper layer boundaries. p. 0 The method was further validated using real field data from active-source and passivewavefield surface wave measurements at the I15 Downhole Array Site. p. 0

Validation and Field Application

The predominant-mode framework was applied to a real field dataset from the I15 Downhole Array Site, where inverted Vs profiles showed strong agreement with invasive downhole PS logs and with empirical transfer functions computed from recorded ground motions. The Rayleigh-wave ellipticity angle extracted from passive-wavefield MAM measurements was found to coincide with the autodetected OP identified by the predominant-mode inversions, providing independent confirmation of the mode interpretation. p. 0 The ETF derived from recorded earthquake ground motions at the site showed good agreement with the TTFs of the best-fitting inverted Vs profiles, confirming that the Vs profiles captured key impedance contrasts controlling seismic site response. p. 0

Conclusion on Robustness

The predominant-mode inversion framework offers a robust alternative to conventional inversion approaches by automatically mitigating mode misidentification and reducing velocity overestimation at sites with strong impedance contrasts. p. 0 p. 0 This approach minimizes artificial distortion of the velocity profile and produces solutions that are generally more accurate than those obtained in prior studies. The method successfully captured major velocity contrasts, respective depths, and the associated osculation point in the dispersion data with good accuracy across synthetic cases. p. 0 The framework is particularly well-suited for situations where the inversion target is constructed from a combination of several MASW active shots with different offsets and/or when combining MASW dispersion data with that obtained from multiple passive MAM arrays. p. 3 The study confirms that removing poorly constrained transition-zone frequencies stabilizes the inversion and enhances the accuracy of the recovered Vs structure for Model k. The predominant-mode inversion successfully captured the target dispersion data well across the usable frequency range, while most of the inversion error is concentrated at a few isolated frequencies. p. 3 This approach yields stable and reliable inversions by avoiding biased fitting at transition frequency bands. p. 8 The study concludes that the predominant-mode inversion framework offers a robust alternative to conventional inversion approaches by automatically mitigating mode misidentification and reducing velocity overestimation at sites with strong impedance contrasts. p. 0 p. 0 The authors also state that the MATLAB code implementing the predominantmode inversion framework will be made publicly available upon publication. The synthetic models used in this study are available through the NHERI DesignSafe Data Repository at https://doi. p. 0 p. 0 p. 0 p. 0 p. 0 p. 0org/10.17603/ds2-cpmr-v194. The field data set is also accessible through DesignSafe data repository at https://doi. p. 0 p. 0 p. 0 p. 0org/10.17603/ds2-xt4m-xa77. The corresponding authors acknowledge the use of the field dataset originally processed and documented by Tyler Jackson as part of his master’s research at Utah State University. p. 0 p. 0 The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. The synthetic models used in this study are available through the NHERI DesignSafe Data Repository at https://doi.org/10.17603/ds2-cpmr-v194.

How the training signal is built

The inversion objective function was defined as a weighted root-mean-square dispersion misfit (md) value between the observed dispersion data and modelled dispersion curves. This misfit is calculated using the formula involving observed phase velocity, modeled predominant-mode velocity, number of frequency points, and standard deviation. p. 3 An additional relative dispersion misfit (mdr) value is required for the predominant-mode inversion framework to identify individual frequency points where the theoretical predominant-mode is unable to adequately fit the target dispersion data. p. 3 The relative dispersion misfit is calculated via a percentage formula involving observed phase velocity, modeled predominant-mode velocity, and standard deviation. p. 3

Where the gains come from

The predominant-mode inversion consistently produces much lower Vs values than the fundamental-mode inversions, confirming that forcing the dispersion data to follow the fundamental mode leads to poorer inversion accuracy. The predominant-mode inversion successfully captures several velocity contrast boundaries, showing good agreement with the true profile, while some deviation occurs at greater depths. The comparison of Vs misfit values demonstrates that the predominant-mode inversion provides a markedly more accurate reconstruction of the Vs profile for Model h than the fundamental-mode inversion. p. 3 The study confirms that the predominant-mode strategy, combined with careful treatment of experimental dispersion data near modal-transition frequencies, yields stable and reliable inversions. p. 3 The comparison of transfer functions shows that the TTFs calculated from the best-fitting inversion-derived profiles align closely with the empirical ground motions recorded at the site, providing independent validation. p. 0

Inversion Procedure Details

The inversion procedure involved global search inversion using Particle Swarm Optimization (PSO) implemented through MATLAB’s built-in Global Optimization Toolbox. The optimization involved multiple unknowns, including layer thicknesses, Vs, and Poisson's ratios. A swarm size of 150 particles was selected to evaluate candidate models in each iteration. p. 3 The inversion objective function is defined as a weighted root-mean-square dispersion misfit (md) value between the observed dispersion data and modelled dispersion curves. p. 1 The optimization process involves multiple stages to balance exploration and exploitation of the model space. p. 0 The final profile with minimum misfit serves as the initial or starting model for subsequent stages. p. 5 The inversion was performed across multiple layering parameterizations to account for inversion layering uncertainty. p. 0 The study utilized a layering ratio (LR) approach, exploring a range of LR values typically from 1.5 to 5. The mass density was fixed at 2000 kg/m3 for all layers, and the Poisson’s ratio was permitted to vary between 0.2 and 0.35. The maximum Vs value was initially limited to 1000 m/s but was increased when ensembles of trial profiles began to reach the limit. p. 1 The inversion procedure involved re-running the inversions after excluding dispersion data with mdr values greater than 5% around the modal transition OP frequency. p. 8 The resulting profiles demonstrate good agreement with the PS-logging data in the upper 100 m, particularly for LR = 1. p. 95 and 2. The theoretical dispersion curves fit the experimental data well and indicate that the predominant-mode inversion captured the target dispersion data well across the usable frequency range. p. 3 The Rayleigh-wave ellipticity angle extracted from passive-wavefield MAM measurements showed a clear shift in particle-motion direction from prograde to retrograde at approximately 1.5 Hz, consistent with the expected ellipticity sign change. p. 6 p. 6 The osculation point identified in the predominant-mode inversion was found to coincide with the frequency range where the Rayleigh-wave particle motion switched from prograde to retrograde, consistent with the expected ellipticity sign change. p. 7 The study confirms that excluding dispersion data near modal transition OP frequency prevents the inversion from attempting to satisfy conflicting modal behavior. p. 0 p. 0 The final results confirm that the predominant-mode strategy, combined with careful treatment of experimental dispersion data near modal-transition frequencies, yields stable and reliable inversions. p. 3 The authors also state that the MATLAB code implementing the predominantmode inversion framework will be made publicly available upon publication. The synthetic models used in this study are available through the NHERI DesignSafe Data Repository at https://doi.org/10.17603/ds2-cpmr-v194. The field data set is also accessible through DesignSafe data repository at https://doi.org/10.17603/ds2-xt4m-xa77.

Summary of Key Findings

The predominant-mode inversion framework successfully captured major velocity contrasts, respective depths, and the associated osculation point in the dispersion data with good accuracy. The predominant-mode inversion consistently produces much lower Vs values than the fundamental-mode inversions. p. 0 The study confirms that forcing the dispersion data to follow the fundamental mode leads to poorer inversion accuracy. p. 0 The predominant-mode inversion was further validated by comparing the ETF derived from recorded earthquake ground motions at the site with the TTFs of the best-fitting inverted Vs profiles, confirming that the Vs profiles captured key impedance contrasts controlling seismic site response. p. 0 The overall misfit distributions confirm that removing poorly constrained transition-zone frequencies stabilizes the inversion and enhances the accuracy of the recovered Vs structure for Model k. The study concludes that the predominant-mode inversion framework offers a robust alternative to conventional inversion approaches by automatically mitigating mode misidentification and reducing velocity overestimation at sites with strong impedance contrasts. The authors also state that the MATLAB code implementing the predominantmode inversion framework will be made publicly available upon publication. The synthetic models used in this study are available through the NHERI DesignSafe Data Repository at https://doi.org/10.17603/ds2-cpmr-v194.

Field Validation

The predominant-mode inversion was applied to a surface wave dataset collected near the I-15 Downhole Array (I15DA) site, where the inverted Vs profiles were shown to agree well with those from invasive Vs measurements. The Rayleigh-wave ellipticity angle extracted from passive-wavefield MAM measurements showed a clear shift in particle-motion direction from prograde to retrograde at approximately 1.5 Hz, consistent with the expected ellipticity sign change. The TTFs calculated from the best-fitting inverted Vs profiles align closely with the ETF resonant frequencies, indicating that the predominant-mode inversion successfully captured the key impedance contrasts in the subsurface. p. 9 These results confirm that the Vs profiles from the predominant-mode inversion captured key impedance contrasts controlling seismic site response. p. 1 The overall misfit distributions confirm that removing poorly constrained transition-zone frequencies stabilizes the inversion and enhances the accuracy of the recovered Vs structure for Model k. The study confirms that excluding dispersion data near modal transition OP frequency prevents the inversion from attempting to satisfy conflicting modal behavior. The predominant-mode inversion framework offers a robust alternative to conventional inversion approaches by automatically mitigating mode misidentification and reducing velocity overestimation at sites with strong impedance contrasts. The authors also state that the MATLAB code implementing the predominantmode inversion framework will be made publicly available upon publication. The synthetic models used in this study are available through the NHERI DesignSafe Data Repository at https://doi.org/10.17603/ds2-cpmr-v194. The field data set is also accessible through DesignSafe data repository at https://doi.org/10.17603/ds2-xt4m-xa77.

Comparison with Fundamental Mode

The fundamental-mode inversion significantly overestimates velocities between roughly 4 and 30 m depth, whereas the predominant-mode inversions successfully capture several velocity contrast boundaries. The fundamental-mode inversion results depart substantially from the true Vs profile at all depths greater than 5 m. p. 8 The predominant-mode inversion yields consistently lower Vs values across all layering ratios, with best-fit values typically below 0. p. 015. The benchmark Vs misfit values reported by Vantassel and Cox (2021) are, in general, slightly higher than the predominant-mode results from this study. The overall misfit distributions confirm that forcing the dispersion data to follow the fundamental mode leads to poorer inversion accuracy. p. 3 The predominant-mode inversion framework offers a robust alternative to conventional inversion approaches by automatically mitigating mode misidentification and reducing velocity overestimation at sites with strong impedance contrasts. The authors also state that the MATLAB code implementing the predominantmode inversion framework will be made publicly available upon publication. The synthetic models used in this study are available through the NHERI DesignSafe Data Repository at https://doi.org/10.17603/ds2-cpmr-v194. The field data set is also accessible through DesignSafe data repository at https://doi.org/10.17603/ds2-xt4m-xa77. The corresponding authors acknowledge the use of the field dataset originally processed and documented by Tyler Jackson as part of his master’s research at Utah State University. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

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. Predominant-mode inversion framework can automatically identify, at each frequency, the Rayleigh-wave mode with the maximum vertical amplitude, allowing inversions to proceed without the need for explicit mode indexing by the inversion analyst.

  2. The system can be used to mitigate errors from mode misidentification by using the mode carrying the theoretical maximum surface energy at each frequency as a basis for inversion, thus avoiding ambiguity in mode numbering made by the inversion analyst.

  3. The framework can be applied when target dispersion data are formed by combining dispersion data from multiple active shots with dispersion data extracted from passive-array recordings that have unknown source locations, which is currently difficult for effective-mode formulations.

  4. The system can be validated by comparing the resulting shear wave velocity misfit values (mV) against benchmarks reported by Vantassel and Cox (2021), showing that the predominant-mode inversion consistently produces much lower mV values than the fundamental-mode inversions.

  5. The framework can be enhanced by excluding dispersion data points with mean relative dispersion misfit (mdr) values exceeding 5% around modal transition OP frequencies, which prevents the inversion from attempting to satisfy conflicting modal behavior and yields more accurate reconstruction of the Vs profile.

  6. The system can provide independent validation by extracting Rayleigh-wave ellipticity angles from passive-wavefield MAM measurements, as the identified osculation point coincides with the frequency range where the Rayleigh-wave particle motion switched from prograde to retrograde, providing independent confirmation of the mode interpretation.

  7. The system can generate theoretically sound transfer functions (TTF) that align closely with empirical transfer functions (ETF) obtained from aftershock ground motions, offering independent validation of the predominant-mode inversion results, beyond the dispersion-data fitting.

Sources

Related papers