Random killing trace and interpolation workflow for 2D seismic line

151 views
Skip to first unread message

Elham Safarzadeh

unread,
Apr 23, 2025, 12:47:52 PMApr 23
to OpendTect Users

Dear All,
Hello,

I’m interested in applying trace interpolation using ML tools in OpendTect, but my dataset is 2D seismic profiles. I tried adapting the 3D tutorial workflow to my 2D profiles, but I wasn't successful.

I have several legacy 2D seismic lines with gaps and no-signal areas, and I’d like to use interpolation and ML to fill these gaps accurately. I’d really appreciate any guidance on  how this can be done with 2D data.

Best regards,
Elham

Paul de Groot

unread,
Apr 23, 2025, 1:01:25 PMApr 23
to us...@opendtect.org
Dear Elham,

OpendTect's pre-trained 3D ML models can be applied to 2D data as well. Please select the "Unet Fill Missing Traces" model from the pre-trained library and press Go. In the application window, select the 2D option.

Best regards,

Paul.

image.png
--
Paul de Groot
Special Adviser


-----------------------------------------------------------------------------------------------------------------------  
dGB Earth Sciences
Phone:+31 53 4315155
E-mail:paul.d...@dgbes.com
Internet:dgbes.com 
----------------------------------------------------------------------------------------------------------------------- 


--
You received this message because you are subscribed to the Google Groups "OpendTect Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to users+un...@opendtect.org.
To view this discussion visit https://groups.google.com/a/opendtect.org/d/msgid/users/257e6908-d645-4c60-8517-900478c1fdcfn%40opendtect.org.

Elham Safarzadeh

unread,
Apr 24, 2025, 7:58:05 AMApr 24
to us...@opendtect.org
Dear Paul

Thank you so much  for your quick response and guidance.
I tried the “Unit Fill Missing Trace” on my 2D profile, it did help and improve the continuity of the weak reflections, but it didn’t fix the gaps or muted areas. As I mentioned before, these are legacy seismic profiles that I digitized, so they’re not real SEGY files. Some of them had interpretation marks that I muted manually.

I also gave the ML tools for seismic image segmentation and regression a try. From what I understood, I need to randomly mute some traces and then train the model. But I got stuck. I wasn’t sure how to mute traces properly.

I followed the tutorial and used the random attribute (like randg(1)) after creating 2D and 3D horizons. Everything was going fine until I reached Step 13 on page 417 of the OpendTect Training Manual (v7.0). Then I couldn’t figure out how to continue.

Regarding the Unet fill trace tool,  the final result  was strange. It has several horizontal missing part at the higher depth. First time I used the default parameters, trace=10 and Z=10 merge mode=blended but in the second test I used trace=30 and Z= 0 and Merge mode= average. both time final result has these horizontal gaps. I attached the screenshot of the muted area of my profile to this email.

Sorry for my long message

Thank you so much for giving me your valuable time

Best regards

Elham

after unet fill trace.pngUnet fill trace.pngOriginal section.pngmuted part.pngoriginal.png

Paul de Groot

unread,
Apr 24, 2025, 10:10:43 AMApr 24
to us...@opendtect.org
Hi Elham,

The "Unet Fill Missing Traces" model was trained to interpolate missing traces whereby missing traces are given as traces with zero values at every sample position. Your case is different. You have narrow steeply dipping gaps with zeroes? (Or are there small values in these gaps?)

I do not know why you get these horizontal bands in the data after you apply the model. Perhaps something is wrong with your data. I'll be happy to have a look at your data if it can be shared. Maybe I can train a dedicated 2D model to interpolate these gaps.

If this is of interest, please send me a backup of your OpendTect project, or one or more representative lines in SEGY format. You can mail this to paul.d...@dgbes.com .

Best regards,

Paul.

--
Paul de Groot
Special Adviser


-----------------------------------------------------------------------------------------------------------------------  
dGB Earth Sciences
Phone:+31 53 4315155
E-mail:paul.d...@dgbes.com
Internet:dgbes.com 
----------------------------------------------------------------------------------------------------------------------- 

Elham Safarzadeh

unread,
Apr 24, 2025, 7:29:04 PMApr 24
to us...@opendtect.org
Dear Paul
Thank you so much for your kind support, I sent it by email. About the horizontal bands. I tried the same test on the lighter profiles with 7-10 MB size. It works correctly but when the size of profiles are big around 20 or 30 MB this horizontal bands appear. I resampled one the profiles from 1ms to 2ms and by decreasing the size the ML tools work correctly. I don't know, is it correct to decrease the sample rate or not. 
Best regards
Elham

Reply all
Reply to author
Forward
0 new messages