I compared inversion results using different model-space parametrizations for early-arrival waveform inversion. Results from isotropic inversion of anisotropic data are similar to the vertical velocity in the shallow parts of the model, and closer to the horizontal velocity model in the deeper part. For joint inversion of vertical velocity and the
parameter, I kept
fixed due to the insensitivity of the data to
changes. A model space parametrized by squares of vertical velocity and horizontal velocity tends to produce updates in vertical velocity and
that have opposite signs, which is unfavorable in practical inversion. The model space parametrized by the logarithm of the squared vertical slowness and epsilon has more reasonable updates and gives far better inversion results. However, ambiguity does exist in the inversion results between vertical velocity and
. The same data misfit can be explained by different combinations of vertical velocity and
, so proper model styling is needed to reduce such ambiguity.

Early-arrival waveform inversion for near-surface velocity and anisotropic parameters: inversion of synthetic data