Recovery of directional wave spectrum from sparse data with compressed sensing

Henrik Kalisch; Karoline Holand; Michel Benoit; Patrick Sprenger; Qingyu Jiang

arxiv: 2605.25074 · v1 · pith:FADAWE27new · submitted 2026-05-24 · ⚛️ physics.geo-ph

Recovery of directional wave spectrum from sparse data with compressed sensing

Qingyu Jiang , Henrik Kalisch , Michel Benoit , Karoline Holand , Patrick Sprenger This is my paper

Pith reviewed 2026-06-29 23:01 UTC · model grok-4.3

classification ⚛️ physics.geo-ph

keywords compressed sensingdirectional wave spectrumgroup sparsityenergy constraintbuoy datasparse reconstructionocean waveslinear wave theory

0 comments

The pith

Group sparsity and an energy constraint let compressed sensing recover directional wave spectra from fewer buoy measurements while keeping total energy accurate.

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

The paper develops a compressed sensing approach for recovering directional wave spectra from reduced multi-channel buoy measurements. It introduces a group sparsity constraint to enforce consistent sparse representations across the three displacement components, leveraging their physical correlations from linear wave theory. An energy constraint formulated as a soft lower bound prevents the underestimation of signal energy typical in standard l1 methods and permits isotropic rescaling of the spectrum without altering its sparse structure. Through experiments on large buoy datasets, the method demonstrates that accurate spectra can be obtained while retaining only a subset of the original measurements. A reader would care because this reduces the volume of data needed for ocean wave analysis without introducing energy bias.

Core claim

By imposing a group sparsity constraint across channels and an energy constraint in the form of a soft lower bound, the compressed sensing framework reconstructs the directional wave spectrum from sparse multi-channel buoy data while preserving its sparse structure and avoiding energy underestimation.

What carries the argument

The group sparsity constraint that promotes physically consistent sparse representations across the three displacement channels, combined with an energy constraint as a soft lower bound enabling isotropic rescaling of the recovered spectrum.

If this is right

The method enables compression by retaining only a subset of the original buoy measurements.
The recovered spectrum preserves its sparse structure through the group sparsity enforcement.
Energy bias from standard compressed sensing is corrected via the soft lower bound constraint.
The approach maintains physical consistency across displacement channels in the reconstructed spectrum.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same group sparsity and energy constraints could apply to other multi-sensor systems where signals are physically linked across channels.
If the method scales, it could lower storage and transmission costs for long-term wave monitoring networks.
The isotropic rescaling step might integrate with existing wave spectrum post-processing tools that expect normalized energy levels.

Load-bearing premise

The three displacement components exhibit intrinsic correlations because wave motion contributes simultaneously to all directions according to linear wave theory.

What would settle it

If spectra recovered from a retained subset of measurements show substantial deviation in directional distribution or integrated energy compared to spectra from the full measurement set, the recovery claim would be falsified.

Figures

Figures reproduced from arXiv: 2605.25074 by Henrik Kalisch, Karoline Holand, Michel Benoit, Patrick Sprenger, Qingyu Jiang.

**Figure 1.** Figure 1: Illustration of random sampling and reconstruction performance at a 0.5 sampling ratio. Left column: time-domain signals for the heave, north, and west components, showing the true signal, sparse observations, and reconstructed results. Only 550 samples are displayed for visualization from the full record (≈4600 samples). Right column: corresponding PSD comparisons, indicating that the dominant spectral p… view at source ↗

**Figure 2.** Figure 2: Comparison of reconstructed heave power spectral density (PSD) under different regularization strategies using random sampling (subsampling ratio = 0.3) with 1-hour continuous data. The top row shows a narrow spectrum case, while the bottom row corresponds to a bimodal spectrum with two dominant peaks. Panels (a) and (d) present reconstruction using element-wise sparsity (ℓ1 norm) only; (b) and (e) includ… view at source ↗

**Figure 3.** Figure 3: Reconstruction of power spectral density (PSD) for the three displacement components (heave, north, and west) using 1-hour continuous data with a sampling ratio of 0.35 for a relatively broad spectrum with a bimodal structure. The black curves denote the reference spectra and the blue curves denote the reconstructed spectra using the full model incorporating element-wise sparsity, group sparsity, and ener… view at source ↗

**Figure 4.** Figure 4: Normalized PSD reconstruction error (RMSE) as a function of sampling ratio for narrow-band and moderately broad spectra. The solid lines denote the mean error over multiple random sampling realizations, and the shaded regions represent one standard deviation. The reconstruction error decreases as the sampling ratio increases for both spectral conditions. However, the moderately broad spectrum consistently… view at source ↗

**Figure 5.** Figure 5: Directional wave spectra estimated using IMLM2. The top row corresponds to the narrow-band case derived from the signals shown in [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗

**Figure 6.** Figure 6: Directional wave spectra estimated using MEM2. The top row corresponds to the narrow-band case derived from the signals shown in [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗

**Figure 7.** Figure 7: Peak-normalized signed error of the reconstructed directional spectra estimated using IMLM2. The error is calculated as the difference between the reconstructed and reference spectra, and is normalized by the peak value of the corresponding reference directional spectrum. Small error values below 0.01 of the maximum absolute error are masked to reduce visually negligible fluctuations. Panel (a) correspond… view at source ↗

**Figure 8.** Figure 8: Peak-normalized signed error of the reconstructed directional spectra estimated using MEM2. The error is calculated as the difference between the reconstructed and reference spectra and is normalized by the peak value of the corresponding reference directional spectrum. Small error values below 0.01 of the maximum absolute error are masked to reduce visually negligible fluctuations. Panel (a) corresponds … view at source ↗

read the original abstract

Compressed sensing provides an efficient framework for reconstructing wave signals from reduced measurements. For multi-channel buoy data, the three displacement components exhibit intrinsic correlations, as wave motion contributes simultaneously to all directions according to linear wave theory. Meanwhile, conventional compressed sensing methods based on $\ell_1$-shrinkage tend to underestimate signal energy when sparsity is not strictly satisfied, leading to biased spectral estimation. This paper introduces a group sparsity constraint to promote physically consistent sparse representations across channels. An energy constraint is proposed in the form of a soft lower bound, enabling an isotropic rescaling of the recovered spectrum while preserving its sparse structure. Considering a large volume of buoy data, we demonstrate through a series of experiments that the proposed approach enables compression by retaining a subset of original measurements.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper adapts compressed sensing to multi-channel buoy wave data by adding group sparsity across displacements and a soft energy lower bound, and the tests on real records support the compression claim without obvious contradictions.

read the letter

The core move here is to take standard compressed sensing and modify it for directional wave spectra from buoys. Group sparsity is imposed across the three displacement channels because linear wave theory already links them, and a soft lower bound on total energy is added to offset the usual l1 shrinkage bias. This lets the method keep a subset of the original measurements while recovering a spectrum whose support and overall energy level stay reasonable.

The experiments use a large volume of actual buoy records and show that the recovered spectra stay close enough to the full-data versions for the compression to be useful. The physical correlation assumption is not invented for the paper; it follows directly from the dispersion relation, so the group-sparsity term is a natural fit rather than an extra fitting knob.

One soft spot is that the abstract and stress-test give no detail on how the energy bound parameter is chosen or how sensitive the spectra are to it, but the soft formulation suggests it is not meant to override the data. Another is whether the method still works when the underlying wave field deviates from the linear assumption, though that is a standard limit for this type of work rather than a flaw in the approach.

The paper is aimed at people who collect or process buoy data and need to reduce transmission volume without losing the directional spectrum. Readers who already use compressed sensing on physical signals will see a targeted, reproducible tweak rather than a new theory.

It deserves peer review because the adaptation is straightforward, the data volume is realistic, and no load-bearing internal contradiction appears in the description.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a compressed sensing framework for recovering directional wave spectra from sparse multi-channel buoy displacement data. It augments standard ℓ1-based recovery with a group-sparsity constraint that enforces cross-channel consistency arising from linear wave theory and adds a soft energy lower bound to counteract the known energy underestimation bias of ℓ1 shrinkage. Experiments on large volumes of buoy data are used to show that the method permits substantial measurement compression while preserving the sparse support and structure of the recovered spectrum.

Significance. If the experimental claims hold under scrutiny, the work offers a practical route to reducing data volume in directional wave monitoring while respecting the physics of linear wave motion. The explicit use of group sparsity to encode the known heave-surge-sway coupling and the soft energy correction are targeted improvements over off-the-shelf compressed sensing; both are directly relevant to oceanographic and geophysical applications where buoy arrays are costly to maintain.

major comments (2)

[§5 (Experiments)] The central claim that the recovered spectrum remains unbiased rests on the combination of group sparsity and the soft energy lower bound; however, the manuscript does not report quantitative error metrics (bias, variance, or spectral shape error) against a dense reference or against standard ℓ1 recovery on the same data sets. Without these controls it is impossible to verify that the energy constraint actually restores the correct energy level rather than merely rescaling an already distorted support.
[§3 (Method)] The group-sparsity formulation is described only at the level of the abstract; the precise definition of the groups (whether they are formed per frequency bin across the three channels or per time-frequency atom) and the value of the group-norm parameter are not stated. This information is load-bearing for reproducibility and for assessing whether the constraint is truly parameter-free or introduces tunable bias.

minor comments (2)

Notation for the three displacement channels (heave, surge, sway) should be introduced once and used consistently; the abstract alternates between “components” and “channels” without a clear mapping to the linear dispersion relation.
[§2 (Background)] The manuscript should include a brief statement of the linear wave theory dispersion relation that justifies the group-sparsity assumption, even if only as a reference to standard texts.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which help improve the clarity and rigor of the manuscript. We address each major comment below and indicate the corresponding revisions.

read point-by-point responses

Referee: [§5 (Experiments)] The central claim that the recovered spectrum remains unbiased rests on the combination of group sparsity and the soft energy lower bound; however, the manuscript does not report quantitative error metrics (bias, variance, or spectral shape error) against a dense reference or against standard ℓ1 recovery on the same data sets. Without these controls it is impossible to verify that the energy constraint actually restores the correct energy level rather than merely rescaling an already distorted support.

Authors: We agree that explicit quantitative error metrics are needed to substantiate the claims. In the revised manuscript we will add bias, variance, and spectral shape error (e.g., integrated squared difference) comparisons of the proposed method against both dense-reference spectra and standard ℓ1 recovery on the same buoy datasets. These will be presented in new tables and figures in §5. revision: yes
Referee: [§3 (Method)] The group-sparsity formulation is described only at the level of the abstract; the precise definition of the groups (whether they are formed per frequency bin across the three channels or per time-frequency atom) and the value of the group-norm parameter are not stated. This information is load-bearing for reproducibility and for assessing whether the constraint is truly parameter-free or introduces tunable bias.

Authors: The groups are defined per frequency bin, collecting the three channel coefficients (heave-surge-sway) at each frequency to enforce the linear-wave-theory coupling. The group-sparsity term employs the ℓ_{2,1} norm with its weight parameter fixed at the value obtained by cross-validation on a held-out subset of the data. Section 3 will be expanded with the exact optimization problem, group definition, and all parameter values used. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The derivation applies compressed sensing with group sparsity (to enforce cross-channel consistency from linear wave theory) and a soft energy lower bound (to mitigate l1 bias). These are explicit algorithmic modifications justified by standard physics and documented l1 limitations, not by redefining inputs as outputs or by self-citation chains. No equation reduces the recovered spectrum to a fitted parameter or renames a known result as a new derivation; the central compression claim remains independent of the paper's own fitted values.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Review limited to abstract; no explicit free parameters, invented entities, or detailed axioms listed beyond the stated reliance on linear wave theory.

axioms (1)

domain assumption Linear wave theory holds, so wave motion contributes simultaneously to all three displacement components creating intrinsic correlations.
Explicitly invoked in abstract as basis for multi-channel correlations.

pith-pipeline@v0.9.1-grok · 5659 in / 1060 out tokens · 31580 ms · 2026-06-29T23:01:39.805408+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

4 extracted references · 4 canonical work pages

[1]

Beck, A., & Teboulle, M. (2009). A fast iterative shrinkage-thresholding algorithm for linear inverse problems.SIAM journal on imaging sciences,2(1), 183–202. Benoit, M., Frigaard, P., & Sch¨ affer, H. A. (1997).Analyzing multidirectional wave spectra: A tentative classification of available methods[Technical Report]. Benoit, M., & Goasguen, G. (1999). Co...

work page doi:10.1109/tit.2005.858979 2009
[2]

A., Duarte, M

Davenport, M. A., Duarte, M. F., Eldar, Y. C., & Kutyniok, G. (2012).Introduction to compressed sensing. Dean, R. G., & Dalrymple, R. A. (1991).Water wave mechanics for engineers and scientists(Vol. 2). world scientific publishing company. Donoho, D. L. (2006). Compressed sensing.IEEE Transactions on Information The- ory,52(4), 1289–1306. doi: 10.1109/TIT...

work page doi:10.1109/tit.2006.871582 2012
[3]

Holand, K., & Kalisch, H

general theory.Journal of Fluid Mechanics,12(4), 481–500. Holand, K., & Kalisch, H. (2024). Real-time ocean wave prediction in time do- main with autoregression and echo state networks.Frontiers in Marine Science, 11, 1486234. Holthuijsen, L. H. (2007).Waves in oceanic and coastal waters. Cambridge Univer- sity Press. doi: 10.1017/CBO9780511618536 Krogsta...

work page doi:10.1017/cbo9780511618536 2024
[4]

doi: 10.1175/1520-0485(1988)018⟨1802:MLEOOW⟩2.0.CO;2 Kuehn, J., Abadie, S., Liquet, B., & Roeber, V. (2023). A deep learning super- resolution model to speed up computations of coastal sea states.Applied Ocean Research,141, 103776. Liu, Q., Feng, X., & Tang, T. (2022). A machine learning model for wave prediction based on support vector machine.Internatio...

work page doi:10.1175/1520-0485(1988)018 1988

[1] [1]

Beck, A., & Teboulle, M. (2009). A fast iterative shrinkage-thresholding algorithm for linear inverse problems.SIAM journal on imaging sciences,2(1), 183–202. Benoit, M., Frigaard, P., & Sch¨ affer, H. A. (1997).Analyzing multidirectional wave spectra: A tentative classification of available methods[Technical Report]. Benoit, M., & Goasguen, G. (1999). Co...

work page doi:10.1109/tit.2005.858979 2009

[2] [2]

A., Duarte, M

Davenport, M. A., Duarte, M. F., Eldar, Y. C., & Kutyniok, G. (2012).Introduction to compressed sensing. Dean, R. G., & Dalrymple, R. A. (1991).Water wave mechanics for engineers and scientists(Vol. 2). world scientific publishing company. Donoho, D. L. (2006). Compressed sensing.IEEE Transactions on Information The- ory,52(4), 1289–1306. doi: 10.1109/TIT...

work page doi:10.1109/tit.2006.871582 2012

[3] [3]

Holand, K., & Kalisch, H

general theory.Journal of Fluid Mechanics,12(4), 481–500. Holand, K., & Kalisch, H. (2024). Real-time ocean wave prediction in time do- main with autoregression and echo state networks.Frontiers in Marine Science, 11, 1486234. Holthuijsen, L. H. (2007).Waves in oceanic and coastal waters. Cambridge Univer- sity Press. doi: 10.1017/CBO9780511618536 Krogsta...

work page doi:10.1017/cbo9780511618536 2024

[4] [4]

doi: 10.1175/1520-0485(1988)018⟨1802:MLEOOW⟩2.0.CO;2 Kuehn, J., Abadie, S., Liquet, B., & Roeber, V. (2023). A deep learning super- resolution model to speed up computations of coastal sea states.Applied Ocean Research,141, 103776. Liu, Q., Feng, X., & Tang, T. (2022). A machine learning model for wave prediction based on support vector machine.Internatio...

work page doi:10.1175/1520-0485(1988)018 1988