f How to trust anisotropy estimates from noisy data - Application to multi-offset VSP

In the framework of inverse theory, the resolution of any model parameter estimate is limited by experimental geometry and by the signal-to-noise ratio. We can only see a filtered version of the true parameters. For each estimated parameter, there exists a filter or resolving kernel which quantifies its resolution. Menke (1984) computed resolving kernels in the case of a cross-hole experiment. Singular value decomposition (SVD) analysis is another method to obtain information on the resolving power of a given geometry (e.g. Pratt & Chapman, 1992). These approaches do not describe parameter resolution as a function of noise in the data. However, this can be achieved when resolving kernels are computed on a regularised inverse problem. The kernels are then functions of a trade-off parameter, itseff selected according to the signal-to-noise ratio.