DAS and ocean waves

It was great to have Ethan introduced the topic at the 2025 WISE meeting in Seattle. With all the questions around the technique and its potential, we can certainly continue some of our discussions here.

For those new to the topic, here are 2 excellent papers:
One of the earliest papers, by Sladen et al.: Distributed sensing of earthquakes and ocean-solid Earth interactions on seafloor telecom cables | Nature Communications
and the one by E. Williams et al., including the very cool use of wave dispersion to get current:
https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2021JC018375

So… Ethan, could you explain to us your understanding of what is it that makes the waves stretch the fibre, and why it should be pressure and not the pressure gradient? (which Shane had convinced me that it made sense )

4 Likes

My answer is in three parts: (1) why the horizontal strain in the seafloor is proportional to pressure, (2) how the strain the cable relates to the strain the seafloor, and (3) why the pressure gradient might fit your data well.

(1) I just re-derived this to make some synthetics the other week, but I think it’s the same as Wayne Crawford’s thesis.

For a homogeneous elastic half-space in 2D, we can write the displacement u in P (\phi) and S (\psi) potentials as u_x = \partial_x\phi - \partial_z\psi and u_z = \partial_z\phi + \partial_x\psi. When inserted into the elastic wave equation, this “Helmholtz decomposition” yields two nominally independent wave equations for P and S waves which are only coupled by the boundary conditions:
\partial_{tt} \phi = \alpha^2 \left( \partial_{xx} + \partial_{zz} \right)\phi and \partial_{tt} \psi = \alpha^2 \left( \partial_{xx} + \partial_{zz} \right)\psi

The boundary condition at the seafloor enforced by a plane harmonic surface wave is that the normal stress at the seafloor must equal the pressure at the bottom \sigma_{zz}(z=0) = p e^{i(\omega t - kx)} where \omega^2=gk\tanh{kh}, in addition to the fact that the shear stress must vanish at the liquid interface \sigma_{xz}(z=0) = 0.

Since the horizontal wavenumber and frequency have been provided for us by the surface wave, it is natural to take the ansatz \phi = \Phi(z) e^{i(\omega t -kx)} and \psi = \Psi(z) e^{i(\omega t -kx)}, yielding:
\Phi'' + \left(\frac{\omega^2}{\alpha^2} - k^2\right) \Phi = 0 and \Psi'' + \left(\frac{\omega^2}{\beta^2} - k^2\right) \Psi = 0

Letting \nu_\alpha^2 = k^2 - \frac{\omega^2}{\alpha^2} and \nu_\beta^2 = k^2 - \frac{\omega^2}{\beta^2} (essentially the vertical wavenumber), the solution is simply
\Phi = a_1 e^{\nu_\alpha z} + a_2 e^{-\nu_\alpha z} (and similarly for \Psi). Because both \nu's are real (\nu_\alpha = \sqrt{k^2-\omega^2/\alpha^2} = k\sqrt{1-c^2/\alpha^2}, where c is the surface wave phase velocity, which is maybe 100x slower than the P or S wave velocities). We know that the correct solution doesn’t increase exponentially into the half-space, so we can assume a_2=0.

We therefore have some linear combination of the P and S waves, which propagates horizontally at the surface wave velocity and decays exponentially into the earth:
\phi = a e^{\nu_\alpha z} e^{i(\omega t -kx)} and \psi = b e^{\nu_\beta z} e^{i(\omega t -kx)}

To evaluate the boundary condition and determine a,b, we can write out the components of stress in terms of the potential functions (letting k_\alpha = \omega/\alpha and similarly with \beta for brevity):
\sigma_{zz} = \mu \left[(2k^2-k_\beta^2) \phi - 2i k\nu_\beta \psi \right]
\sigma_{xz} = \mu \left[-2ik\nu_\alpha \phi + (k_\beta^2-2k^2)\psi \right]

Factoring out the oscillatory term and evaluating this at z=0, then solving the resulting system of equations, one obtains
a = \frac{2k^2-k_\beta^2}{\mu R(k)} p and b = \frac{2ik\nu_\alpha}{\mu R(k)} p where R(k) = (2k^2 -k_\beta^2)^2 - 4k^2\nu_\alpha\nu_\beta is Rayleigh’s function and p is the wave pressure at the seafloor.

We can write the horizontal displacement in terms of the potentials:
u_x = \partial_x\phi - \partial_z\psi = -ik\phi - \nu_\beta\psi

Evaluating this at the seafloor (z=0, so e^{\nu_\alpha z} = 1):
u_x = \left[ \frac{k_\beta^2-2k^2 - 2 \nu_\alpha\nu_\beta}{\mu R(k)} \right] ikp e^{i(\omega t - kx)}
\varepsilon_{xx} = \partial_x u_x = -ik u_x = \left[ \frac{ k_\beta^2-2k^2 - 2\nu_\alpha\nu_\beta }{\mu R(k)} \right] k^2 p e^{i(\omega t - kx)}

Recalling the pressure boundary condition \sigma_{zz}(z=0) = p e^{i(\omega t - kx)}, you will note that the horizontal strain is in phase with the pressure. The pressure gradient associated with this surface wave is -ikp e^{i(\omega t - kx)}, which is 90 degrees out of phase from the strain. (Also notably, the opposite is true if you go evaluate the vertical displacement and strain – the vertical displacement is in phase with the pressure, which may be the source of confusion).

I will have to write the second and third parts later!

2 Likes

An addendum to part (1): the other important aspect of this solution is the scale invariance of the ratio between pressure and strain (for a homogeneous earth).

Using the various wavenumber definitions and factoring out k^4 on the top and bottom, the non-oscillatory part of the strain can be written:
\varepsilon_{xx} = \left[ \frac{ \frac{c^2}{\beta^2} - 2\frac{c}{\beta}\frac{c}{\alpha} - 2 }{\frac{c^4}{\beta^4}- 4\frac{c^2}{\beta^2} - 4\frac{c}{\alpha} \frac{c}{\beta} + 2} \right] \frac{p}{\mu}

The ratios c/\alpha and c/\beta are all very small and I think Crawford uses a taylor expansion to obtain a nicer form of the result (I believe you also can eliminate the c dependence that way). Regardless, the point is that the ratio between horizontal strain and pressure is a dimensionless constant. Repeating the same analysis with pressure gradient will yield an extra k.

(2) How does the strain in the cable relate to the strain in the seafloor? Given the large variability among calibration coefficients from some of the work we saw yesterday, I would suggest that we don’t really know – but an idealized model never hurts. One approach is to view the cable as an infinite perfectly cylindrical elastic rod embedded in a uniform elastic medium (the seafloor). This has been considered by Kuvshinov, though he did not discuss the elements of his solutions which are relevant to this community. Brad Lipovsky and I have been working on expanding the elastic rod model to multi-layered cables and various wave types… but that is still work in progress.

I can summarize some of the main points:

  • If the sediment has finite shear rigidity and the cable is frictionally coupled to the sediment, then the axial strain in the cable is going to be identical to the strain in the same direction in the sediment. This is because of the condition that the axial displacement must be continuous across the “welded” cable-sediment boundary. Since the displacement is a harmonic function, its derivative (strain) is also going to be continuous across the boundary. Therefore we expect the proportionality between strain and pressure derived above to hold for the DAS-measured strain.
  • If the sediment has no shear rigidity (a fluid) or the cable is freely slipping, then there is no condition on the axial displacement or shear stress across the cable-sediment interface – only (like in the compliance problem) that the pressure in the sediment must equal the radial stress in the cable.
  • In this latter case, the strain in the cable comes entirely from the Poisson effect, i.e. you squeeze a cylinder radially and it extends axially. To consider this effect, we must first think of two end-member stress states: (a) an infinitely long cylinder with axially-invariant radial pressure applied is in plane strain, meaning that there is no axial strain regardless of the Poisson’s ratio of the cable material, (b) a thin cylinder (relative to the radius) with a axially-invariant radial pressure extends freely according to \varepsilon_{rr} = -\nu \varepsilon_{zz} (where here z is the cable axis and \nu is Poisson’s ratio, unlike the previous derivation).
  • However, the forcing from a surface wave is not axially uniform. It has some apparent wavelength along the cable axis (unless it is perfectly perpendicular). Consequently, the axial strain in the slipping cable case lies somewhere in between the limiting cases (a) and (b) – and where on the continuum it lies is a function of the wavelength, which determines how axially confined the cylinder is and therefore how much axial strain occurs for a given radial pressure.
  • This last point introduces the possibility that the cable response could be proportional to \partial_x p insomuch as \partial_x p spits out an additional factor k—and the sign is right, the shorter the wavelength the less confined the cable, the larger the axial response to constant pressure.
  • I find this model very difficult to accept because the response in the Poisson-only limit is going to be very small relative the response of the coupled cable (possibly 2-3 orders of magnitude smaller), at which point you have to start worrying about the 2nd order aspects of DAS response including photoelasticity (see Kuvshinov). More likely in the fluid case (which we don’t expect for buried cables), things like cable curvature or partial frictional coupling take it out of plane strain so that the pressure can induce axial strain independent of wavelength.

(3) Why does your data look like it is proportional to pressure gradient (other than the far-fetched explanation in 2)? One reason might be the frequency dependence of compliance. Introducing an extra factor of k by taking \partial_x p rescales the pressure data to suppress long wavelengths and enhance short wavelengths. The compliance of the seafloor (as derived in 1) is constant for a uniform half-space, but if the elastic moduli of sediments increases with depth (as in reality), then the compliance increases with wavenumber. The elastic deformation forced by a short wavelength surface wave has a correspondingly short vertical length scale 1/\nu in the solid earth, and is therefore sensitive to moduli of the shallower sediments, which are softer. Therefore, multiplying by an extra k might just be accidentally compensating for the wavelength-dependence of the transfer function.

A second reason is that using \partial_x p instead of p introduces a factor of \cos\theta, which (somewhat) approximates the DAS directional response. As derived in (1), the compliant deformation of the seafloor under a surface wave is vertically polarized, meaning that the strain \varepsilon_{xx} for a wave propagating in the \hat{x} direction is the only component of horizontal strain. If you have a cable in the \hat{x}' direction at an angle \theta to \hat{x}, then the strain along the cable by simple tensor rotation is \varepsilon_{x'x'} = \varepsilon_{xx} \cos^2\theta. If the pressure at the seafloor is p = p_0 e^{i(\omega t-kx)} then the derivative of the pressure in the cable direction is \partial_{x'}p = p k\cos\theta. For a distribution of waves from different directions, this \cos\theta mutes the broadside incident waves – not as severely as the ideal \cos^2\theta, but probably enough to fit the data much better than the non-directionally-corrected pressure. Given you have wavenumber resolution in the SWOT data, it makes sense to introduce a \cos^2\theta scaling to the pressure before comparing it with the DAS data.

1 Like