Number of sample greater than 100

This forum is dedicated to discuss all problems and suggestions related to the inversion software
Posts: 4
Joined: Tue Jun 26, 2012 9:35 am

Number of sample greater than 100

Postby Syl » Mon Jul 18, 2016 9:35 pm

Hi Marc,

is there a way to turn off the following warning when running dinver in command line ?

"The effective number of samples for dispersion curve is greater than 100 (113). This is likely to slow down the inversion process. A usual number of samples is 50. A high number of samples may be observed when various curves are inverted together (dispersion, spac,...). In this case make sure that all these curves use the same frequency samples. Resample all curves if necessary."


Site Admin
Posts: 424
Joined: Mon Aug 13, 2007 11:48 am
Location: ISTerre

Re: Number of sample greater than 100

Postby admin » Mon May 29, 2017 9:43 am

Hi Sylvain,

Do you really need more than 100 points to define your dispersion curve? Usually with 50 it is enough. If you have several curves to invert together and if they do not have the same sampling, the number of total samples can be high. The most annoying consequence is to slow down the inversion because the dispersion curves must be computed for a lot distinct frequencies.

The common 'good' practice is to resample all curves with the same sampling, usually starting a bit lower than the resonance frequency of the site and a bit after the highest available frequency. There are various reasons to extend the frequency range:
  1. The stability of the forward computation is better if the dispersion curves are computed on a large frequency range. Internal checks to detect mode jumping are more efficient.
  2. It lets you check on the inversion results if the chosen parametrization has a sufficient number of degrees of freedom. With a small number of parameters, the model extrapolate the velocity down to great depths, eventually without any limit of resolution. You will probably see on the dispersion plot that the velocity at low frequency is just an extrapolation of the experimental curve. A well behaved parametrization should produce an ensemble of dispersion curves which dispersion explode outside the experimentally constrained data frequencies.
  3. There is a flag 'enable/disable' for each sample. All extrapolated samples generated during the resampling are set to 'disable' by default. They are not included in the misfit computation and they have absolutely no influence over the inversion results.

If you really need those samples, you will have the warning once after Dinver startup. You can check/uncheck the option at the bottom of the dialog box to get rid of the message for the next created run.

Return to “Dinver usage”

Who is online

Users browsing this forum: No registered users and 1 guest