SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsdisba

disba

Data & analytics

View repository

Updated 10-16 02:00

About this skill

disba: layered surface-wave dispersion

Use disba for 1D isotropic layered-model dispersion and sensitivity kernels. It is a forward solver; extracting observed dispersion from raw waveforms or performing tomography requires separate processing and inversion design.

Model and period conventions

  • Supply four equal-length one-dimensional arrays: thickness in km, Vp/Vs in km/s, and density in g/cm³. A row-per-layer table must be transposed once: PhaseDispersion(*model.T).
  • Do not use *zip(thickness, vp, vs, rho): with four layers it silently interchanges properties, and other layer counts give the wrong arguments.
  • Treat the last layer as the half-space and write its thickness as 0.0. Other layers have positive thickness. The final thickness is not a finite bottom boundary, even if a nonzero value is accepted by the library.
  • Use finite positive periods in increasing order. Density must be positive; for the solid models below require vs > 0 and vp**2 > (4/3) * vs**2. Preserve measured units and model provenance.

Phase and group velocity

This complete example uses a solid layered model. The first call may take longer while Numba compiles the solver.

import numpy as np
from disba import PhaseDispersion, GroupDispersion

thickness = np.array([0.5, 1.0, 2.0, 0.0])  # km
vp = np.array([1.5, 2.5, 4.0, 6.0])          # km/s
vs = np.array([0.8, 1.4, 2.3, 3.5])          # km/s
rho = np.array([1.8, 2.0, 2.3, 2.6])         # g/cm³
periods = np.linspace(0.1, 5.0, 50)          # s, increasing

phase_solver = PhaseDispersion(thickness, vp, vs, rho)
group_solver = GroupDispersion(thickness, vp, vs, rho)
rayleigh = phase_solver(periods, mode=0, wave='rayleigh')
love = phase_solver(periods, mode=0, wave='love')
group = group_solver(periods, mode=0, wave='rayleigh')
period_s, phase_km_s = rayleigh.period, rayleigh.velocity

Each result is a DispersionCurve named tuple, not a velocity array. Use result.period and result.velocity together: unavailable roots may be omitted, especially for higher modes. len(result) counts tuple fields, not computed periods; use result.period.size.

Read dispersion curves for plotting, phase/group conversion, mode tracking and frequency conversion. Love waves depend on Vs, density and layer thickness in an isotropic model, while Vp does not enter the SH problem. Modal depth sensitivity is model dependent.

Sensitivity kernels

Continue from the model above. The call takes a scalar period as t or its first positional argument; period= is not a supported keyword.

from disba import PhaseSensitivity

sensitivity = PhaseSensitivity(thickness, vp, vs, rho)
kernel_vs = sensitivity(1.0, mode=0, wave='rayleigh', parameter='velocity_s')
depth_km, dc_dvs = kernel_vs.depth, kernel_vs.kernel

Other parameters are velocity_p, density, and thickness. These are numerical derivatives in the model's units. Check a kernel against a small finite perturbation before using it in an inverse problem; one kernel does not establish unique depth resolution.

Prepare models for parameter studies

Read velocity model preparation for shape, unit and physical checks, table transposition, and an explicit forward-model function. Supply density and Vp estimates deliberately; do not silently turn a Vs profile into an uncalibrated density model. The solid-model validator in that reference does not cover water layers; fluid-layer and Love-wave boundary conditions require separate treatment.

For incomplete curves, record the mode, wave type, returned period range and solver settings. Catch disba.DispersionError only when handling a failed root search, and retain the diagnostic. Do not suppress arbitrary exceptions or relabel malformed inputs as nonexistent modes.

Verification and resources

Examples were checked with disba 0.7.0 and NumPy 1.26.4 on 2026-09-14 using synthetic solid models. Checks cover a homogeneous Rayleigh limit, layered phase/group curves, numerical sensitivities and the helper's CSV/plot output and failure diagnostics. They do not validate field-data inversion or fluid-layer models.