gphistogram have an anomalous grapfic output

Post in this forum all topics related to all sesarray softwares, not a particular one.
Post Reply
luigiV
Posts: 15
Joined: Fri Jul 17, 2020 9:28 am

gphistogram have an anomalous grapfic output

Post by luigiV »

Hi Marc,
I'm a Geopsy user since the first release (2005..or so..).
Now I have installed 3 realease (2.9.0., snapshot -20161121 and win64 3.3.0) on two laptops.
It happens with the last release the graphic output of RTBF on gphistogram shows
the slowness frequency graph (or speed) where the speeds are higher at high frequencies than at low frequencies. Am I making any mistakes in setting the parameters or it is a bug in the program? With previous releases it doesn't happen using max2curve (which isn't in this release).
The database is composed with 5 stations 3C and the records are 2h long.

Another thing, SPAC does not work: just clicked on the icon the program crashes (the same database on previous releases reaches the end of the processing without problems).
I attache a screenshot of gphistogram issue.

Thanks a lot for the anwer and best regards
Luigi Veronese
Attachments
gpHistogram.jpg
(132.39 KiB) Not downloaded yet

admin
Site Admin
Posts: 490
Joined: Mon Aug 13, 2007 11:48 am
Location: ISTerre
Contact:

Re: gphistogram have an anomalous grapfic output

Post by admin »

Hi Luigi.

Thans for you fidelity!
On you plot the lower limit of the coloured area is due to the GRID_SIZE parameter. If you increase the grid size, the results will be extended to higher frequency and lower velocity. But, for this case, I'm unsure that you will get more results. Can you plot the theoretical array limits computed with warangps on top of your plot? (copy and paste layers from the dispersion plot obtained in warangps).
Did you tried gpviewmax? It is still preliminary release but it helps to plot more results (azimuth, ellipticity,...)

SPAC is now fixed in release 3.3.1

luigiV
Posts: 15
Joined: Fri Jul 17, 2020 9:28 am

Re: gphistogram have an anomalous grapfic output

Post by luigiV »

May be I unsterstod the problem. The five stations have a diamond shape array, but the distance between the stations is perhaps very large (200 m, too much?) for F-K analysis and may be the
wave sources are no right on azimuth. In fact I have experimented on a database of years ago acquired for different purposes and now I am waiting to be able to examine more suitable data for radii and number of stations.
Have a good day
Last edited by luigiV on Wed Jul 29, 2020 5:14 pm, edited 1 time in total.

luigiV
Posts: 15
Joined: Fri Jul 17, 2020 9:28 am

Re: gphistogram have an anomalous grapfic output

Post by luigiV »

Sorry, I forgot your suggestion to use warangs.
I did it and I send you the image of the dispersion curves, station coordinates and the values ​​of kmin and kmx.
I hope to have rightly understod your suggestion
warangs_result.jpg
(90.83 KiB) Not downloaded yet

luigiV
Posts: 15
Joined: Fri Jul 17, 2020 9:28 am

Re: gphistogram have an anomalous grapfic output

Post by luigiV »

Hi Marc,
waiting for you to find time to read my reruns, I thought I'd read all your papers (I had already read your thesis) and tried to process the data, for which I had made the first post regarding gphistogram, with the SPAC. I realized that the RTBF could not work due to the large distance between the stations and perhaps the azimuth distribution of the sources was not adequate. I would therefore like to present the results obtained, well aware that the layout and the number of stations are not adequate. Using a previous release (2.9.0.) I was able to open the spac2disp tool (which unfortunately does not allow to define the grid step). I have defined only one ring and unfortunately the dispersion plot is not well clear. I have chosen (correctly I hope) the dispersion curve that covers a very small frequency range (0.5-2Hz) given the large radius. With dinver I first inverted the autocorrelation curve assuming a simple model (one layer on a half-space) and without constrains for the Vs and a maximum depth of 200m. I have prepared a document containing all the results of the three hypotheses (inversion of the autocorrelation curve, joint inversion with DC for the 2-layer and 3-layer model). The difference in terms of misfit is relatively small, but that of the profile of Vs consistent and above all there is no fitting between synthetic DC and the SPAC DC one.
I do not have "a priori" certain data on the Vs profile and bedrock depth, but the profiles obtained by the joint HVDC inversion of all five stations (the DC curves were obtained from ReMi array of 250m in length and 24 geophones). The site is located within the area studied by C. Mascandola et al. (BSSA) "Mapping the seismic bedrock of the Po Plain (Italy) trough Ambient-Vibration Monitoring". The resonant frequency of the site is 0.65-0.7Hz and the estimated depth of the bedrock at 160m is very similar to that of the 2-layer SPAC DC model. What is your opinion about my SPAC data approach and
spac.pdf
(997.09 KiB) Downloaded 49 times
the insufficient fitting between DC SPAC and synthetic DC? I would also like to ask you something about the Joint Inversion HVDC with Dinver which in my opinion is not satisfactory, but I will do it on another occasion.
I sincerely hope you have the time and desire to answer me.
Thank you
Luigi

admin
Site Admin
Posts: 490
Joined: Mon Aug 13, 2007 11:48 am
Location: ISTerre
Contact:

Re: gphistogram have an anomalous grapfic output

Post by admin »

Hi Luigi,

I'm fine with your spac curve and the selection you did in spac2disp. Just make sure that kmin does not correspond to wavelength too large compared to your array aperture. Roughly speaking, your array has a 500 m aperture. A kmin of 0.0015 corresponds to a wavelength of 4189 m (lambda=2*pi/k) which is more than 8 times the array aperture. Usually SPAC can get reliable results down to wavelength 2-3 times the array aperture. Some authors even claim for 6 to 7 times but to stay on the safe side and without other information it is better to limits the autocorr curve at low frequency. So I would limit kmin to 0.005 (wavelength of 1250 m)

Then, I do not understand the origin of the DC curves plotted (red and blue) on the third line of plots, on the right with the comment "DC-DC edited". The first inversion results are OK. The try with 3 layer is also fine but I do not understand your DC curve in black. Is it a synthetic curve? Is it related to experimental data? Does it come from the REMI dataset?

As you know that the resonance frequency is around 0.65-0.7 Hz, you can check that the collection of obtained profiles fits with this frequency. Use a terminal, for instance the one provided by GeopsyLand:

Code: Select all

cd /to/your/dinver/directory/where/your/run.report/has/been/saved
gpdcreport run_01.report -best 1 | gpell | figue -c
This plots the ellipticity curve of the Rayleigh fundamental mode for the best model. The peak corresponds approximately to the H/V peak if the structure is correct. This process is useful for complex 3-layer models. For simple 2-layer models the simple formula (Vs/(4*H)) gives 0.58 Hz in your case. It is not far from 0.65 Hz. Probably that inverting with 3 layers and adding the HV peak frequency constraint will provide good results.

luigiV
Posts: 15
Joined: Fri Jul 17, 2020 9:28 am

Re: gphistogram have an anomalous grapfic output

Post by luigiV »

Thanks Marc.
The DC curve is the one saved as a multicolumn file with spac2disp together with the * .target used for inversion. The edited DC curve is the same but smoothed. Is it therefore a wrong operation to jointly invert the autocorrelation curve and its corresponding DC? What appears in black in the plot is this. However, it has slowness values ​​typical of bedrock and therefore what you make me observe when you say that kmin is too large is correct. So what does that dispersion curve mean?
The dispersion curve ReMi is very different in slowness (typical of soils) and stops at 1Hz (60-70m in depth) but the picking is always uncertain (personal choise) at lower frequencies.
I followed your suggestion to check the ellipticity corresponding to the best 2 layers Vs profile and I was puzzled because I was used, with the previous Geopsy releases, to see the H/V curves expressed in amplitude and not in degrees. It seems to me that the curve is consistent with the experimental data since the (negative) peak is at 0.65Hz. I am attaching the relative image. I then tried to jointly invert the autocorrelation curve with one of the experimental H/V. The misfit is not excellent, but the 4 layers Vs model puts the bedrock at 160-180m in good agreement with my previous results. However, how can I verify the fitting of the experimental H/V curve with the calculated ones if only synthetic ones appear in the plot? Is it not possible because the H/V curve refers to 3C data and the synthetic one only to 1C data (so plotted in a negative specular fashion)?
Thanks again for your patience in helping me understand and in giving me the precious suggestions that are not found in the online documentation.
Attachments
4L.pdf
(644.8 KiB) Downloaded 51 times

luigiV
Posts: 15
Joined: Fri Jul 17, 2020 9:28 am

Re: gphistogram have an anomalous grapfic output

Post by luigiV »

Sorry, A last little question about inversion. In your answer to my previuos post you say: "Probably that inverting with 3 layers and adding the HV peak frequency constraint will provide good results"
How can I add the HV peak frequency constraint? Adding the H/V curve to autocorr curve or....

Thanks

admin
Site Admin
Posts: 490
Joined: Mon Aug 13, 2007 11:48 am
Location: ISTerre
Contact:

Re: gphistogram have an anomalous grapfic output

Post by admin »

In dinver target, select "Peak frequency", and just enter the value.
You will see that playing with the relative weight of sub-targets is necessary to achieve a reasonable fit of the dispersion curve. An alternative and better way is to set a minimum misfit for all sub-targets. Once the minimum misfit is reached for a sub-target, the fit cannot be improved for this sub-target. The only way to get a better fit is to improve all sub-target fits. This way you will get acceptable models for the autocorr curve and for the peak frequency. The difficulty comes from the fact that fitting a single value is much easier than fitting a whole curve.

admin
Site Admin
Posts: 490
Joined: Mon Aug 13, 2007 11:48 am
Location: ISTerre
Contact:

Re: gphistogram have an anomalous grapfic output

Post by admin »

Answers to your message posted on 02 Aug 3:32 pm.

If you derived a dispersion curve from the autocorr curve, it is useless to invert the autocorr and the dispersion. You are free to choose one of the two options but not both at the same time. Both objects contain the same information. If I remember well there were discrepancies between the dispersion curves calculated during the inversion of autocorr and the black dispersion curve. Are you sure that there is no error while transforming the DC? (not excluding spac2disp export error to be eventuelly investigated).

Options available in gpell can produce the usual positive output (-abs). Be extremely careful when inverting an ellipticity. If you provide the classical H/V curve, it contains Love, Rayleigh and body waves. The calculated ellipticity is for a pure Rayleigh wave. Hence you assume that Love wave do not contribute to the wavefield, which is a bit unrealistic.Inverted Vp and Vs profiles may be dramatically biased by error on the ellipticity amplitude. In ellipticity target, for each curve, you can specify if it is signed or absolute values.

A conservative option is to consider only the frequency of the peak. That's what I suggested. There are also ways to extract the Rayleigh ellipticity from a single station measurement:

luigiV
Posts: 15
Joined: Fri Jul 17, 2020 9:28 am

Re: gphistogram have an anomalous grapfic output

Post by luigiV »

Hi Marc,
first of all sorry for the silly question about how to enter the peak frequency data in Dinver (a quick inspection made me immediately understand that I shouldn't have done it ... after all the years I used Dinver).
What puts me in trouble is, if I understand correctly, you suggest to be very careful in the joint inversion of the HV and DC curves as the first contains all types of waves and the second only Rayleigh or Love. In fact when I did this with Dinver I always found that the best fitting was for the H/V curve, while it was extremely rare to be able to obtain an acceptable fitting with the DC. In many recent papers I have read that the HVDC joint inversion is desirable to obtain a correct estimate of the bedrock position and that this processing provides a solution to the non-uniqueness problem of minima. And that is precisely the essential support of the DC for identifying the true solution among the many minima provided by the inversion of the HV curve alone.
You understood that I am referring to the DFA and the HVinv software that I have recently used with personal satisfaction, noting that greater emphasis the best fit is reserved to the DC curve. I know that you have also collaborated to some extent on this project and therefore I would like to hear your opinion on this.
About the DC curve derived with Spac2disp I confirm that the speed values ​​were exactly those (typical of bedrock).
Last thing: the Dunand method I have often used to confirm that the pulse around the maximum peak H/V is natural and not anthropogenic. Now I read with pleasure that it will soon become a tool for extracting Rayleigh's ellipticity from the 3C time series.
Thanks a lot and best regards
Luigi

p.s.
The Geopsyland minishell doesn't accept the command -abs (as you suggest for a positive graph)
I tried to use the HVTFA tool with your data.
Questions: how many time needs to complete the process? I have been forced to stop because it doensn't finish in a resonable time and the output .max file have a different format from the one presented in Neries technical paper
(# File generated by Geopsy, H/V TFA processing
# seconds from start | cfreq | H/V | AmpZ | AmpH | Delay
9.6 2.00135 0.747509 2169.77 1621.92
9.6 2.00135 0.807737 2169.77 1752.6
21.92 2.00135 1.50149 911.714 1368.93
21.92 2.00135 1.46191 911.714 1332.84
31.17 2.00135 1.07907 2073.86 2237.83
31.17 2.00135 1.19933 2073.86 2487.24
35.76 2.00135 0.503696 1660.05 836.158
35.76 2.00135 0.44642 1660.05 741.079
40.38 2.00135 7.90909 310.292 2454.13
40.38 2.00135 7.99398 310.292 2480.47....)
and is not possibile to open it with max2curve (crash) nor gpviewmax (doesn't start after launch).
I found in your database 3 bug reports concerning the same problem with Windows 10 (2714, 3715 and 3860) and which apparently have never been fixed. I found that using the snapshot 0.0.0. the TFA tool runs and closes the file autonomously, while the 3.3.0-3.3.2 releases do not and I got a 1.6Gb .max file ...!
I found a temporary solution when I found out that max2curve from release 2.9.0. (which does not have the HVTFA tool) correctly opens the max files generated with the 0.0.0 snapshot. At first I tried to replace the max2curve of release 3.3.2. with that one of 2.9.0. (adding the necessary dlls), but it didn't work. What happened with the snapshot instead, as you can see from the attached image. However, I believe it is necessary to fix the bug with the newer releases on Windows.

In my previous post I asked you why in the plot of calculated H/V curves I cannot graphically compare those with the measured one because it does not appear in the plot (the degrees are positive and those of the theoretical curves negative). What is the reason?
Attachments
HVTFA snapshot.jpg
(178.54 KiB) Not downloaded yet
4L_mod.pdf
(304.69 KiB) Downloaded 51 times
Last edited by luigiV on Sun Aug 09, 2020 12:46 pm, edited 6 times in total.

Post Reply