Curve sampling

Re-sampling and cutting are often used for real datasets to get a uniform distribution of samples along the frequency axis. This document describes different techniques (from the simplest to more advanced ones) to define the curve sampling.

Re-sampling is also strongly advised when dealing with several modes (or curves in general). Computing a higher mode dispersion curve for a sample set always requires the computation of all the lower modes for the same sample set. Hence, to reduce the computational load, it is better to have the same sampling for all curves. If curves are defined over distinct frequency ranges, the validity of each sample can toggled to avoid a misfit computation at a particular frequency.

Validity of sample points

The distribution of samples influences the inversion results by changing the weights put on some frequency ranges. For instance, a uniform sampling in frequency puts a higher weight on high frequencies whilst a uniform sampling in period puts a higher weight on low frequencies. A log sampling is an intermediate position.