Page 1 of 1

Tuning parameters - joint inversion

Posted: Mon Jan 30, 2017 10:50 am
by fmpg

I am inverting the dispersion and ellipticity curves (joint inversion) of a determined soil profile. I am inverting the theoretical curves (dispersion curve for f<3Hz and left and right flank of the elliptitity curve) in order to understand the best way to get to the right velocity profile.

After performing several inversions considering different values of tuning parameters (mainly these combinations):
i) It = 500, Ns0=500, Ns = 500, Nr = 1000 (Nr>Ns for a more exploratory search, right?),
ii) It = 600, Ns0=500, Ns = 350, Nr = 200 (in general have been leading to lower misfit values)
I am obtaining the same tendencies of velocity models (in most of the, the interface between soil and bedrock is around 30m depth, when it should be at 37m depth).
Only in one of those runs (run 13, in attachment) (using param- ii)), I obtained velocity profiles similar to the right one.

I checked the Parameter space of different runs and it seemed that, although in the beginnig there it covers a wide range of values, at a certain point, DVs03 tends to 30m.

- Is this a problem of entrapment in local minima?
- Do you recommend other values for tuning parameters in order to perform an optimum exploration of the space and avoid this?

In order to force that result, I reduced the range of DVs03 to 34-100m, in the inversion. However
I obtained the same values (DVs03 around 30m, even after setting Dvs03>34m) (run_37_prof.png).

- Why this happened?

Best regards,
Fátima Gouveia

Tuning parameters associated to the results in attachment:
- run_13: It = 600, Ns0=500, Ns = 350, Nr = 200 (best fitted model) [210500 total models] (min. misfit=0.0138)
- run_17: It = 200, Ns0=200, Ns = 1000, Nr = 1000 (considered Ns=Nr for exploration phase) [200200 total models] (min. misfit=0.04515)
- run 18: It = 200, Ns0=200, Ns = 500, Nr = 1000 (200 initial models, 500 models generated in 1000 best cells, right?) [100200 total models] (min. misfit=0.06537)