Characterizing Ocean Flows with the Scattering Transform

2025

Read Characterizing Ocean Flows with the Scattering Transform (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-00819.txt.

Introduction

Skinner and co-authors read ocean surface maps with a tool that sees shape, not just strength. Power spectra can say how much motion sits at each scale. They cannot always tell a wave from a swirl when the strengths match. The scattering transform, built from wavelet folds, pulls out geometric clues the spectrum misses.

On model sea height fields, the tool split regions by their hidden dynamics. Balanced flows, internal waves, and kinds of turbulence each left a distinct mark, even at equal spectra. The authors offer the frame for reading satellite height maps and radar scans. They pitch it for fluid puzzles well past the ocean.

Catalog record

YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-00819

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Characterizing Ocean Flows with the Scattering Transform Abstract † † journal: Geophysical Research Letters † † authors: J. W. Skinner 1,2 , A. Lawrence 2 , J. Callies 1 † † corresponding: J. W. Skinner, [email protected] \justifying \addeditor JS \addeditor AL California Institute of Technology, Pasadena, CA Brandeis University, Waltham, MA keypoints The scattering transform is a data analysis method that can extract and quantify the morphology and distribution of features in fields. It can distinguish ocean flows that have significant geometric differences in physical space but may have the same power spectra. This can be used to identify waves and different flavors of geophysical turbulence in satellite altimetry and other snapshot observations. Abstract Upper-ocean flows are a multi-scale jigsaw puzzle of turbulence and waves. Characterizing these flows is essential for understanding their role in redistributing heat, carbon, and nutrients, yet power spectral analysis cannot always distinguish between types of motion. We show that the scattering transform (ST), a wavelet convolution method, can extract geometric information from flow fields, offering insights beyond the power spectrum. The ST distinguishes balanced dynamics, internal waves, and types of turbulence—even when their power spectra are identical. Applied to sea surface height (SSH) fields from ocean models, the ST differentiates regions with distinct underlying dynamics. Our analysis offers a framework for interpreting SSH from satellite altimetry missions and for analyzing other spatial maps (e.g., from airborne and coastal radar). More generally, the ST is an appealing way to characterize complex fluid motion in a variety of geophysical contexts. Plain Language Summary Ocean currents take a range of different shapes: swift narrow currents along temperature fronts, swirls akin to weather systems in the atmosphere (albeit much smaller), and wavy motions excited by winds and tides. It is not always easy to distinguish between these types of currents in observations and numerical simulations, but how they transport heat, carbon, and nutrients in the ocean is very different. While waves reversibly deform tracers, different types of turbulent motions irreversibly stir and mix them. Here, we describe how a new analysis technique, called the scattering transform, can characterize the geometry of flow structures and thereby offer important clues toward the underlying dynamics of observed and simulated ocean currents. 1 Introduction The upper ocean is a complex jigsaw puzzle of intertwined flows spanning a wide range of spatial and temporal scales. At the mesoscale ( 100300\text{~}\mathrm{km} ), eddies and jets advect heat, salt, carbon, and other tracers, shaping the large-scale circulation and biogeochemistry ( Wunsch (2007) ; Morrow & Le Traon (2012) , e.g.,) . At the submesoscale ( 1100\text{~}\mathrm{km} ), fronts, filaments, and eddies drive vertical nutrient fluxes and carbon sequestration, which may be critical for sustaining marine ecosystems and regulating atmospheric {\rm CO}_{2} ( Klein & Lapeyre (2009) ; Mahadevan (2016) ; McWilliams (2016) ; Lévy et al. (2012) ; Lévy et al. (2018) ; Taylor & Thompson (2023) , e.g.,) . Internal waves, generated primarily by winds and tides, propagate through this dynamic system of currents, redistributing energy, shearing stratified layers, and facilitating small-scale mixing that impacts global climate ( Jochum et al. (2013) ; Whalen et al. (2020) , e.g.,) . Disentangling these different flows is crucial for understanding their individual contributions to transport, predicting oceanic responses to climate change, and improving ocean models. This remains a significant challenge, particularly for submesoscale flows, because spatially resolved observations often lack the temporal resolution needed to separate turbulence and waves. The power spectrum is an important tool for characterizing different flow regimes and their variability (e.g., Ferrari & Wunsch (2010) ; Callies & Ferrari (2013) ; Callies et al. (2015) ; Chereskin et al. (2019) ; Lawrence & Callies (2022) , among many others) . Many ocean flows are practically indistinguishable in their power spectra, however, because they follow the same power law. For example, a forward energy cascade in three-dimensional turbulence ( Kolmogorov (1941) ) , an inverse energy cascade in two-dimensional turbulence ( Kraichnan (1967) ) , and a forward cascade of surface potential energy in surface quasi-geostrophic turbulence ( Blumen (1978) ) all predict a k^{-5/3} power law for the kinetic-energy spectrum ( k is the horizontal wavenumber). In practice, this is also often indistinguishable from the empirical Garrett & Munk (1972) model for the internal-wave field, which has a k^{-2} roll-off. Tools beyond the power spectrum are therefore needed to characterize the distinct morphology of these flows. Strongly nonlinear dynamics organize turbulent upper-ocean flows into eddies and filaments ( Batchelor (1953) ; Reeves (1969) ; Townsend (1976) ; Gargett (1989) ; Frisch (1995) ; Thorpe et al. (2007) ; Callies et al. (2015) ) , whose spatial localization and coherence is encoded in the phase of the Fourier coefficients, which are discarded by the power spectrum ( Armi & Flament (1985) ; Oppenheim & Lim (1981) ) . Figure 1: Building intuition for the scattering transform. (a) Comparison on structural information captured by the power spectrum, bispectrum, and scattering transform. A turbulent vorticity field (left panel) with spatially intermittent and geometrically complex multi-scale structures is reconstructed using the information captured by each method. The power spectrum reconstruction randomizes the Fourier phases of the turbulent field; the bispectrum and scattering reconstructions optimize Gaussian random fields via gradient descent to match the respective statistics of the turbulent field ( Cheng & Ménard (2021) , following) . (b) Illustration of the diagnostics \mathrm{s}_{22} (shape) and \mathrm{s}_{21} (sparsity). The turbulent field (middle panel) is modified to generate new fields where all \mathrm{s}_{22} or \mathrm{s}_{21} coefficients are modulated by \pm 25\% . The scattering transform (ST) ( Mallat (2012) ) retains significantly more structural information than the power spectrum, making it effective for characterizing flow morphology (Fig. 1 a). While the bispectrum, which measures interactions between different wavenumber components, captures some structural information, it is still less effective than the ST ( Cheng et al. (2024) ) . The ST captures geometric properties—e.g., localization, distribution, elongation, and orientation—through wavelet convolutions and modulus operations ( Cheng & Ménard (2021) ) . The ST quantifies this information as a function of spatial scale, so dynamics at different scales can be probed. In this letter, we apply the ST to idealized models that capture balanced flows, waves, and various turbulent regimes as well as to realistic ocean simulations that include internal gravity waves and meso- and submesoscale turbulence. We specifically focus on characterizing flow structures in the sea surface height (SSH) field with an eye to the high-resolution spatial maps provided by the Surface Water and Ocean Topography (SWOT) mission ( Fu et al. (2009) ; Morrow et al. (2019) ; Fu et al. (2024) ) . 2 The Scattering Transform The scattering transform (ST) is a representation of an input field I=I(\mathbf{x}) , where \mathbf{x}=(x,y) , obtained by combining wavelet transforms, modulus operations, and spatial averaging ( Mallat (2012) ; Bruna & Mallat (2013) ) . The first- and second-order scattering coefficients are \displaystyle S_{1}(j_{1},l_{1}) \displaystyle=\langle|I\star\chi^{j_{1},l_{1}}|\rangle, (1) \displaystyle S_{2}(j_{1},j_{2},l_{1},l_{2}) \displaystyle=\langle||I\star\chi^{j_{1},l_{1}}|\star\chi^{j_{2},l_{2}}|\rangle, (2) where \star denotes the convolution, \left|\,\cdot\,\right| the modulus, \langle\,\cdot\,\rangle the spatial average, and j_{i} and l_{i} are the spatial scale 2^{j_{i}} (in pixels) and orientation of the wavelet \chi^{j_{i},l_{i}} used in the i th convolution ( Mallat (2012) ; Cheng

Read the complete text