{"id":"dfa08220-5c73-40b5-a07f-f7a17e1e141e","arxiv_id":"2505.00819","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"The scattering transform distinguishes balanced flows, internal waves, and two turbulence types in ocean models, even when their power spectra are identical.","lead":"This paper tests whether a wavelet-based analysis tool called the scattering transform can tell apart different types of ocean flows that look the same in standard power-spectrum measurements. It shows the method can separate waves from swirling currents and different kinds of turbulence, and it may help interpret sea-surface height maps from satellite missions like SWOT.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'even when their power spectra are identical' claim rests on the balanced-vs-IGW experiment, but spectra are only matched at initialization, not verified at the analysis time t=1000 when the ST is applied.","rationale":"The reader's weakest_assumption precisely identifies the most load-bearing weakness: spectral matching is only asserted at initialization. My reading of Section 4 confirms that the time series in Fig. 3c shows the balanced and IGW cases diverging in s21/s22, and the text attributes this to nonlinear balanced evolution, which is expected to alter the power spectrum. Since the headline claim is explicitly about identical power spectra, this is not a minor technicality. The 2D-vs-SQG experiment is unaffected because spectra are shown at equilibrium; this should be credited. The realistic section is suggestive but depends on an Eulerian filter and is not the keystone of the 'identical spectra' claim. If the proposed test shows the t=1000 spectra differ, the appropriate fix is to weaken the abstract to 'even when initialized with identical power spectra' or to re-run the experiment under maintained spectral matching (e.g., forcing/dissipation that pins E(k)). This does not overturn the paper; the ST method and the 2D-vs-SQG demonstration remain valuable. Therefore the conditional verdict is appropriate and I recommend no change.","tokens_in":13377,"tokens_out":4951,"duration_ms":56511,"concrete_test":"Post-process the t=1000 ∇²h fields from the 20-member balanced and IGW ensembles (scripts are public) to compute isotropic power spectra E(k) per ensemble; quantify separation with a two-sample test or a spectral-distance metric across the j1=2–5 bands used for s21/s22. If the spectra are statistically indistinguishable, the claim stands. If they differ, rerun the ST analysis on phase-randomized surrogates that are constrained to have the identical spectrum at t=1000; if the s21/s22 separation collapses, the 'even when identical spectra' conclusion must be weakened.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that the scattering transform distinguishes balanced flows, internal waves, and turbulence 'even when their power spectra are identical.' This is rigorously demonstrated for 2D vs SQG turbulence, where Fig. 2(i) shows 20-member kinetic-energy spectra at equilibrium. The same cannot be said for the balanced-vs-IGW shallow-water experiment in Section 4. Simulations are initialized with matched E(k) ∝ k^{-2} spectra, but the ST is applied to ∇²h at t=1000, after the balanced flow has undergone an inverse cascade, vortex mergers, and filament sharpening while the IGW field has dispersed and smoothed. These nonlinear processes almost certainly reshape E(k), yet no spectra at t=1000 are shown. Since s21 and s22 are scale-dependent ratios of band-passed statistics, a spectral rearrangement between the two ensembles could produce differences in s21/s22 even if the geometric information were absent. The abstract's 'even when their power spectra are identical' is therefore only verified at initialization for this case; at the analysis time the claim is unsubstantiated. The omission is directly relevant because the balanced-vs-IGW test is the one closest to the satellite-altimetry application.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes the scattering transform (ST) as a diagnostic for characterizing the geometry of upper-ocean flows beyond what can be learned from power spectra. It defines two scale-dependent summary statistics—s21 (sparsity) and s22 (shape)—and shows, in three experimental settings, that they separate dynamical regimes: (i) 2D inverse-cascade turbulence versus surface quasi-geostrophic (SQG) forward-cascade turbulence, both configured to have kinetic-energy spectra with k^{-5/3} slopes; (ii) balanced turbulence versus inertia-gravity waves in freely decaying shallow-water equations initialized with identical k^{-2} kinetic-energy spectra; and (iii) two regions of a realistic North Atlantic simulation with different balances between mesoscale/submesoscale turbulence and internal waves. The paper argues that the ST extracts phase/geometry information that the power spectrum discards, and it positions the method as a framework for interpreting SWOT sea-surface-height snapshots. The code and packages used are publicly available.","tokens_in":13612,"tokens_out":14512,"duration_ms":126092,"significance":"The central claim—that the ST can distinguish flows even when their power spectra are identical—is of considerable interest for oceanography, because many observed flows share similar spectral slopes. The 2D-versus-SQG experiment in Section 3 is well posed: the analyzed ∇²ψ fields share the same power-law spectrum, the ensemble statistics show small error bars, and the differences in s21 and s22 are consistent with the visual morphology. The realistic North Atlantic application uses an independent temporal filter as a ground-truth check, which is a sound validation strategy. The paper is clearly written and uses open-source, reproducible tools, which are strengths. However, the strongest claim is not fully supported for the balanced-versus-IGW experiment in Section 4, where the spectra are matched only at initialization and not verified at the analysis time; this is the case closest to the satellite-altimetry motivation. The manuscript needs to address this gap or carefully qualify its claims.","major_comments":[{"comment":"The abstract claims that the ST distinguishes balanced flows and internal waves 'even when their power spectra are identical,' but the balanced-versus-IGW shallow-water experiment does not verify that the two ensembles have identical power spectra at the time the ST is applied. The simulations are initialized with matched E(k) ∝ k^{-2} spectra, yet the ST is applied to ∇²h at t=1000, after the balanced flow has undergone an inverse cascade and filament sharpening while the IGW field has dispersed and smoothed through weak nonlinear interactions. The manuscript does not show the power spectra of the analyzed ∇²h fields at t=1000 or at the times displayed in Fig. 3c. Because s21 and s22 are scale-dependent statistics, a spectral rearrangement alone could produce differences between the ensembles, so the claimed geometric discrimination is not established for this experiment. Please provide the ensemble-mean spectra at the analysis time, or perform a controlled test with spectra matched at the analysis time, and adjust the claims accordingly. If the spectra differ at t=1000, restrict the 'identical power spectra' claim to the 2D-versus-SQG experiment or phrase it more cautiously.","section":"Section 4, Fig. 3"},{"comment":"The abstract states that the ST distinguishes 'types of turbulence—even when their power spectra are identical,' but Section 3 describes the 2D and SQG simulations as configured to equilibrate 'with identical spectral slopes in their inertial ranges,' not with identical power spectra. The analyzed ∇²ψ fields will then have the same power-law slope, but the text does not state whether the full spectra (including prefactor and dissipation-range structure) coincide over the analyzed scales. Since s21 and s22 are amplitude-ratio statistics, a global amplitude difference would not matter, but differences in the shape of the spectrum (e.g., inertial-range boundaries) could still influence the scale-dependent statistics. Please either show that the power spectra of the analyzed fields are identical in the analyzed range, or rephrase the claim to 'even when their power spectra have the same slope' or 'similar power spectra.' This distinction is directly relevant to the paper's headline claim.","section":"Section 3 and Abstract"}],"minor_comments":[{"comment":"The text contains a typo: 'Guassian random fields' should read 'Gaussian random fields.'","section":"Section 7.3"},{"comment":"The model start date '06-12-2025' appears to be a future date relative to the manuscript submission; please clarify whether this is a model time or a calendar date, and if it is a typo, correct it.","section":"Section 7.4"},{"comment":"In the SQG case, the dynamically active field is the surface buoyancy ξ = |∇|ψ, while the text and figure caption refer to the analyzed field as 'the vorticity field ∇²ψ' for both 2D and SQG turbulence. The analysis of ∇²ψ for SQG is a legitimate choice, but the terminology should be explained to avoid confusion.","section":"Section 3 and Figure 2"},{"comment":"The definition of s22(j1,j2) = ⟨S∥2/S⊥2⟩_{l1} would be clearer if the text explicitly stated that for each l1 the average is over second-order coefficients whose wavelet orientation l2 is parallel (∥) or perpendicular (⊥) to l1.","section":"Equation (4)"},{"comment":"A brief discussion of the robustness of the ST summary statistics to measurement noise would be useful, since SWOT observations will contain significant noise and the Laplacian ∇²h amplifies small-scale noise.","section":"Section 5"}],"recommendation":"major_revision","confidential_remarks":"The manuscript fits the scope of GRL and the scattering transform is a promising and potentially impactful diagnostic for ocean altimetry. The two major comments concern the degree to which the headline claim ('even when their power spectra are identical') is supported by the experiments. In particular, the balanced-versus-IGW experiment in Section 4 does not verify spectral identity at the analysis time, which is the case most relevant to the application. I believe these issues are addressable within the manuscript's scope, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The scattering transform is a good fit for this problem, and the paper shows it convincingly for the case where it matters most. The 2D-vs-SQG experiment is the heart of the paper, and it is done carefully: the two turbulence flavors are configured to equilibrate with identical kinetic-energy spectra, analyzed on the same vorticity fields, and the s22/s21 differences are large compared to the ensemble scatter. That is a real, reproducible result and it makes the case that the ST captures geometric information the power spectrum throws away. The realistic North Atlantic application is also a sensible test, and the seasonal progression in the ENA versus WNA is qualitatively what you'd expect from wave versus balanced dominance. Credit is due for using 20-member ensembles, for checking resolution insensitivity, and for making the code public.\n\nThe soft spot is the one the abstract leans on hardest. \"Even when their power spectra are identical\" is rigorously demonstrated only for the 2D-vs-SQG comparison. In the shallow-water experiment, the balanced and IGW fields start with matched k^-2 spectra, but the ST is applied at t = 1000 after the balanced flow has undergone an inverse cascade, vortex mergers, and filament sharpening, while the IGW field has dispersed and smoothed. The paper does not show the spectra at the analysis time, so the claim that the ST separates balanced flow from waves under identical spectra is unverified for that case. The time series in Fig. 3c actually suggest the two cases evolve differently, so the likely resolution is that the spectra have changed and the claim needs to be softened or the spectra shown. That is a moderate fix, not a fatal flaw. Also, there is an unfilled citation placeholder in Section 4 that needs to be removed or completed.\n\nThe Eulerian filter in the realistic section is acknowledged as a simplification; given the scale of the question, that is acceptable for a GRL letter, though it limits how much weight you can put on the full-vs-subinertial comparison.\n\nWho is this for? Anyone working with SWOT SSH snapshots or with model evaluation who needs a shape-sensitive diagnostic beyond spectra. It will likely be cited as the oceanographic scattering-transform reference. It deserves a serious referee: the analysis is mostly sound, the method is clearly explained, and the main weakness is a missing spectral check rather than a conceptual error. I would send it to review and ask the authors to verify the spectral matching at the analysis time or qualify the abstract claim.","headline":"Useful new application of the scattering transform to ocean flows, with the strongest claim (identical spectra) fully checked only for the 2D-vs-SQG case.","tokens_in":14158,"tokens_out":1492,"would_cite":true,"duration_ms":17121,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper argues that the scattering transform, a wavelet-based method, can separate ocean flows—balanced currents, internal waves, and turbulence types—even when their power spectra are identical.","keywords":["scattering transform","sea surface height","internal waves","balanced dynamics","geostrophic turbulence","submesoscale","satellite altimetry","power spectrum"],"falsifier":"Recompute the isotropic power spectra of the $\\nabla^2 h$ fields at $t=1000$ for both the balanced and inertia-gravity-wave ensembles and compare them; if they have diverged, the 'identical power spectra' claim for that case is not established. A stronger test would phase-randomize one balanced snapshot to force identical spectra and check whether the ST still separates the fields.","tokens_in":13152,"feed_emoji":"🌊","tokens_out":9055,"duration_ms":80215,"temperature":0.7,"pith_summary":"This paper tries to establish that the scattering transform (ST), a wavelet-based method that records the shape and arrangement of structures, can characterize upper-ocean flows beyond what the power spectrum provides. The ST converts a spatial field into coefficients that quantify how sparse and how elongated its features are at each scale. In idealized simulations the authors show that two turbulence types with the same $k^{-5/3}$ spectra—2D turbulence and surface quasi-geostrophic turbulence—are separable by these statistics, and that balanced flow and inertia-gravity waves separate as well. Applied to sea surface height from a realistic North Atlantic simulation, the ST separates a wave-dominated eastern region from a turbulence-dominated western region and tracks the seasonal transition between regimes. If this holds for satellite snapshots, it would let altimetry missions like SWOT identify underlying ocean dynamics without the temporal filtering that current approaches require.","feed_headline":"Scattering transform distinguishes ocean flows with identical spectra","feed_subtitle":"Two geometry metrics, sparsity and shape, tell waves from turbulence in single snapshots.","key_machinery":"The central object is the scattering transform: convolve the field with a family of rotated and dilated Morlet wavelets (plane waves modulated by Gaussian envelopes), take the modulus, convolve again, and spatially average. First-order coefficients capture scale-by-scale amplitudes; second-order coefficients capture how structures at different scales and orientations interact. The paper reduces this to two orientation-averaged ratios: $s_{21}$ (sparsity) and $s_{22}$ (shape). It is these ratios, computed across scale pairs $j_1, j_2$, that carry the discrimination: they quantify the localization and anisotropy that distinguish filaments, eddies, and wave patterns, while remaining insensitive to the Fourier phases that the power spectrum ignores.","core_discovery":"The paper's central claim is that morphology, not just spectral power, identifies oceanic flow regimes. Two summary scattering statistics—sparsity $s_{21}$, which measures how localized or clustered features are, and shape $s_{22}$, which measures whether structures are smooth or elongated—separate dynamical classes. For equal power spectra, the scattering coefficients remain distinct because they retain phase and spatial information that the power spectrum discards. The paper demonstrates this for 2D versus SQG turbulence with matched $E(k)\\propto k^{-5/3}$ spectra, and for balanced shallow-water flow versus inertia-gravity waves initialized from the same Garrett-Munk spectrum. In realistic sea surface height fields the statistics separate western and eastern North Atlantic regions and reveal a winter-to-summer shift toward wave-dominated signals.","pith_inferences":["One consequence the paper leaves implicit is that the ST statistics could serve as an unsupervised diagnostic for satellite swath data, flagging regions where internal tides contaminate the balanced signal without requiring harmonic fits to time series.","A direct extension, not tested here, would be to check whether $s_{21}$ and $s_{22}$ remain robust to realistic altimeter noise, swath gaps, and the anisotropic sampling of SWOT.","Because sparsity and shape vary with scale, the same machinery could be applied to subsurface fields such as buoyancy or tracer concentrations to compare stirring geometry across models and observations."],"forward_implications":["Single snapshots of sea surface height can be classified by dynamical regime, making temporally sparse altimetry such as SWOT useful for separating waves from balanced motion.","Model evaluation gains a stricter test: simulations must reproduce not only spectral slopes but also the geometry of fronts, filaments, eddies, and waves.","Because the statistics are scale-resolved, they identify the wavenumber range over which waves or balanced flow dominates, as shown for 4–8 km versus 32–64 km scales.","The method transfers to other spatial maps, such as airborne or coastal radar, and to other geophysical fields.","Seasonal and regional contrasts in upper-ocean dynamics can be tracked from daily snapshots without temporal filtering."],"supporting_citations":[{"why":"Defines the scattering transform and its invariance properties, providing the mathematical foundation for the method.","marker":"(Mallat, 2012)"},{"why":"Introduces the sparsity and shape statistics $s_{21}$ and $s_{22}$ and the field-reconstruction framework used to interpret them.","marker":"(Cheng & Ménard, 2021)"},{"why":"Compares the scattering transform with the bispectrum and supports the claim that the ST captures more structural information.","marker":"(Cheng, Morel, Allys, Ménard, & Mallat, 2024)"},{"why":"Provides the $k^{-5/3}$ inverse-cascade spectrum for two-dimensional turbulence, one of the matched-spectra baselines.","marker":"(Kraichnan, 1967)"},{"why":"Provides the $k^{-5/3}$ forward-cascade spectrum for surface quasi-geostrophic turbulence, the other matched-spectra baseline.","marker":"(Blumen, 1978)"},{"why":"Establishes the SQG dynamics whose sharp filaments and vortex morphology the ST separates.","marker":"(Held et al., 1995)"},{"why":"Supplies the internal-wave spectral shape used to initialize the shallow-water wave-versus-balanced experiments.","marker":"(Garrett & Munk, 1972)"},{"why":"Modifies the shallow-water equations to prevent nonlinear steepening of gravity waves, isolating the wave dynamics under study.","marker":"(Bühler, 1998)"},{"why":"Supplies the realistic North Atlantic simulation whose full and subinertial sea surface height fields are analyzed.","marker":"(Sinha, Callies, & Menemenlis, 2023)"}],"fun_headline_variants":["Wave vs turbulence: scattering transform sees what spectra miss","Scattering transform tells ocean flows apart when spectra match","Two scattering metrics unmask ocean flow regimes","Sparsity and shape: new keys to ocean dynamics","Scattering transform goes beyond power spectra for ocean flows"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that, in the wave-versus-balanced shallow-water experiment, the two flow types still have identical power spectra when the scattering transform is applied; the initial spectra are matched, but the spectra at analysis time are not shown.","fun_headline_variants_meta":{"raw":{"variants":["Wave vs turbulence: scattering transform sees what spectra miss","Scattering transform tells ocean flows apart when spectra match","Two scattering metrics unmask ocean flow regimes","Sparsity and shape: new keys to ocean dynamics","Scattering transform goes beyond power spectra for ocean flows"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000458,"raw_usage":{"total_tokens":2241,"prompt_tokens":837,"completion_tokens":1404,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":453,"completion_tokens_details":{"reasoning_tokens":1328}},"tokens_in":453,"tokens_out":1404,"duration_ms":10236,"temperature":1.0,"reasoning_tokens":1328,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T04:34:02.919962+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the isotropic power spectra of the $\\nabla^2 h$ fields at $t=1000$ for both the balanced and inertia-gravity-wave ensembles and compare them; if they have diverged, the 'identical power spectra' claim for that case is not established. A stronger test would phase-randomize one balanced snapshot to force identical spectra and check whether the ST still separates the fields.","supporting_citations":[{"cited_title":", Morel, R","cited_arxiv_id":null,"evidence_quote":"Compares the scattering transform with the bispectrum and supports the claim that the ST captures more structural information."},{"cited_title":"APACrefauthors \\ 1967","cited_arxiv_id":null,"evidence_quote":"Provides the $k^{-5/3}$ inverse-cascade spectrum for two-dimensional turbulence, one of the matched-spectra baselines."},{"cited_title":", Pierrehumbert, R T","cited_arxiv_id":null,"evidence_quote":"Establishes the SQG dynamics whose sharp filaments and vortex morphology the ST separates."},{"cited_title":"\\ Munk, W","cited_arxiv_id":null,"evidence_quote":"Supplies the internal-wave spectral shape used to initialize the shallow-water wave-versus-balanced experiments."},{"cited_title":", Callies , J","cited_arxiv_id":null,"evidence_quote":"Supplies the realistic North Atlantic simulation whose full and subinertial sea surface height fields are analyzed."}],"review_version":1}