Best software for seismic refraction tomography

[InspecteurColombin]

I'll re-process these data as well.
I have a question concerning the way you draw the travel-time curves. I am more used to the way exposed below:

Travel-time curve

What do you do it that like this? To help to check if the arrival times are OK and easily find picking errors?

[kaminae]

Yes, I use same type of graphics (without "0" times) (x=(sx+rx)/2) . It's comfortable to control reciprocity times.
And in addition (x=rx) to control hodographs shape.

[InspecteurColombin]

OK, thanks for the explanation.
Why don't you use zero times? To avoid bad triggering conditions and possible time shifts?
When surveying with a sledge hammer, I usually shoot close to a geophone so that I can set it to 0 when editing and picking. When shooting with explosives, this isn't a problem anymore since the trigger is accurate (controlled either by the cutting of a copper cable or by GPS clocks).
Using zero time seems important to me since it allows to check for reciprocity (e.g. if the time shift is different between two shots, how can reciprocity be assessed?).

Please find below the updated data inverted with the SIRT algorithm. Grid cell size of 2.5*2.5 m, homogeneous starting model (Vp=4000 m/s; which can be deduced from a quick refraction analysis) and final RMS error = 0.12% after 15 iterations.

Velocity model with rays

[kaminae]

1. Of course I take special attention to start moments.
For taking account of "start time" I use some options (from reciprocity times or calculated data) in my picker (in manual or automatic mode) and also I can invert start times with velocities together.

2. Next step real data or synthetic data with strong topography?

3. I would like to know what ray tracing algorithm is used for forward promblem calculation?
And what inversion procedure You used for first yesterday pictures(You wrote it takes many times ).

[kaminae]

It seems to me Your software used Moser algorithm for ray tracing with smooth velocity distribution(linear interpolation inside cell).

[john09]

Both of your results look great! Anyway, I would be satisfied if I can get such images from field surveys. But I do have a question about locating underground air-filled voids. Is seismic refraction tomography good at locating such kind of voids? I suspect they might cause confusion in the algorithm because they don't let seismic waves passing through them at all.

[InspecteurColombin]

2. Next step real data or synthetic data with strong topography?

I'm OK for real data with strong topography. The only data set I have with varying topography is not mine so I cannot post it on a forum...

3. I would like to know what ray tracing algorithm is used for forward promblem calculation?

I don't know.

And what inversion procedure You used for first yesterday pictures(You wrote it takes many times ).

This is a simulated annealing algorithm implemented in Seisopt. Informations & peer-reviewed papers here : http://www.optimsoftware.com/. I ran it yesterday with a medium resolution but it did not give satisfactory results. So I ran it again today with higher resolution. And it is still running.

[InspecteurColombin]

It seems to me Your software used Moser algorithm for ray tracing with smooth velocity distribution(linear interpolation inside cell).

I do not know about moser, but yes, the problem is linearized within each cell.

[Preston09]

Both of your results look great! Anyway, I would be satisfied if I can get such images from field surveys. But I do have a question about locating underground air-filled voids. Is seismic refraction tomography good at locating such kind of voids? I suspect they might cause confusion in the algorithm because they don't let seismic waves passing through them at all.

I think electrical resistivity tomography is much better at locating air-fiiled voids.

[InspecteurColombin]

Here are the results of the inversion of Kaminae's data set using the Seisopt@2d code. The grid is made of 126*88 cells. Cell size is 0.8*0.4 m, horizontally and vertically, respectively. Computation for this particular solution took 60 mn. As I tested 10 different model sizes, the total computation time is about 10 hours. The final error, in a RMS sense, between observed and calculated first break picks is 4%.

[InspecteurColombin]

Both of your results look great! Anyway, I would be satisfied if I can get such images from field surveys. But I do have a question about locating underground air-filled voids. Is seismic refraction tomography good at locating such kind of voids? I suspect they might cause confusion in the algorithm because they don't let seismic waves passing through them at all.

I agree with preston09. Seismic refraction tomography might not be the most suitable technique for detecting underground air-filled cavities. A couple of years ago, we conducted a test over a cavity (3 m in diameter at a depth of 5 m) within limestones. We tested ERT, seismic refraction tomography, EM31, GPR and MASW. Results:

• ERT is not efficient to detect a resistive anomaly within a resistive environment (that is not very new...). It might have workd if the cavity was filled with water and/or clay ;
• Seismic tomography did not work as well ;
• Because of a conductive top soil layer, GPR did not work ;
• EM31 did not work because the depth to the target was too important and also because the technique is not suitable to this particular geological setting. It might have worked if the resistive anomaly (void) was located within a conductive environment ;
• Only MASW was successful. Bu we used a special approach by calculating the spectral ratios between transmitted over incident waves. Doing this for different and adjacent couples of geophones (with same spacing) allowed us to produce a longitudinal profile of spectral ratios. When strong attenuation took place, we interpreted this as being a result of frequency filtering due to the cavity. We validated this approach to locate the location and depth of fissures within clays. (this approach is not genuine as well, it was first tested and validated, both numerically and experimentally, to locate fissures within concrete)

[kaminae]

Sorry for silence. Was at business trip.
There is field data set with strong topography.
Alex

[InspecteurColombin]

Kaminae,
I started giving a look at your data. I think there might be a problem with the first four shots (please see figure below: pink, black, red and blue curves). Curves are very regular (is it the reality?) and picked times do not show a decrease towards the shot position (filled stars at the base of the time axis).

In the same way, some other curves do not show a minimum time at shot location (see shot at 210m).
Finally, at these same shot locations, picks regularly show values of 10 to 15 ms. Were source located along or nearby the profile?

[kaminae]

I can't answered - it's not mine data and not mine picking.

[Preston09]

Finally I got some time and I tried SeisImager with the SAGEEP data set. Unfortunately, although I tried all kinds of parameters and settings in the program (it needs manual input of some parameters for inversion), it always crashed without getting any results. My conclusion is that SeisImager is not good for seismic refraction tomography. It's wonderful to see your programs working so well.

[InspecteurColombin]

Hello Preston,
You might invert the other data set we provided so that we will be able to compare seisimager with the rest.
I am going to propose real data with varying (but small) topography today or tomorrow (Vp and Vs).

[Preston09]

Results from Kamine's "data.txt" using seisImager. Running time is about 10 minutes.

[InspecteurColombin]

Hello all,
Sorry Kaminae but I did not manage to obtain satisfying results with your data. If you are OK, you might find here real data with little topography (irregular elevation of 20 m along a 115 m-long profile). Distances are along the ground and not horizontal (i.e. not projected). 2 data set are provided: P-waves and S-waves which, interestingly, do not provide the same results. Data come from a clay landslide. Integrated analysis of P and S-waves allow to find a water table and (normally) 2 shear surfaces.

p-waves.txt

(5.95 KiB) Downloaded 581 times

s-waves.txt

(5.55 KiB) Downloaded 571 times

Enjoy

[Preston09]

Inversion of InspecteurColombin's data "seismic3layers_fault.txt" using SeisImager2D:

[Preston09]

InspecteurColombin, here are Inversion results from your data p-waves.txt:

Result:

Cells:

Cells and Raytracing:

[Preston09]

I changed the parameters (mainly the targeted depth) and I think I got better results:

RayTracing:

[InspecteurColombin]

Preston, do you have a way to estimate/quantify an error on the final model?

Here is what I got with Seisopt.
Grid is made of 57*61 cells of size 2*1.3 m horizontally and vertically, respectively.
Final error, in a RMS sense, is 4.9%
Hits which mimic a ray-tracing indicate an interface at a depth of 4 to 6 m. This corresponds to the water table. Since geophone spacing is 5 m, this depth is not very well resolved/imaged.

[kaminae]

Results with zondst2d:
Vp and Vs

Vs model is more deep.

final RMS
1.5% for smooth inversion
4% for layered inversion

I had some problems with Vp Vs joined inversion for layered model(V - differ, boundaries same) because final RMS for joined inversion > 8%.

[Preston09]

The RMS errors in percentage sense are also around 5% in my results.

[ncoord]

So there are no conclusions about which one is the best? All the programs can play out the raytracing, show the grid, and give confidence of errors. The all look working for me.