-
Notifications
You must be signed in to change notification settings - Fork 29
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
I tried FWI with JUDI on the actual marine data, but my gradient was always 0 #176
Comments
First of all are you sure that your gradient is all zero?
|
|
Is it possible that there is a problem with my coordinate setting? In the seismic data,I set the y-coordinate of the source point, the receiver point and the CMP point to 0. The X-coordinate of my source,receiver and CMP points do not coincide. And I set model geometry by:
|
From the first look I don't see any mistakes.
Also I would pay attention if
And also what returns Saving these results to file will get you understanding what is wrong. And for testing purposes you don't need to run all the shots, probably 1 shot at each iteration would be enough for debugging purposes. |
Thank you for your advice Sir. I will check it as you said |
@zhangxiaoshuotttt yes the problem may be related to the coordinates. You have to check coordinate scaler trace header as well. To be sure that you coordinates are fine please do the forward modeling for a single shot with the same model. If your coordinates are completely wrong then forward modeling will return empty. I see that More over JUDI options for FWI should have |
I think there is a scaling issue accessing th le min x through the headers as it should be rescaled by 1e4. |
You can double check by making sure the origin coincides with |
|
I rezeroed the origin by uniformly subtracting the x origin coordinate. Next, I will make a forward model to verify whether there are still problems with my coordinates. Thank you for your advice. |
My "Function val" is not zero , it is "3.47902e+07". |
I think your wavelet has a time shift so there is very little illumination. Check what JUDI default tucker wavelet looks like you'll see it need to start around time=0 |
Thank you very much, this advice is really helpful! |
Discussed in #175
Originally posted by zhangxiaoshuotttt March 17, 2023
Hi,
I tried to obtain FWI results from marine data, but I can't get gradient, it is always 0.
I would appreciate some help or tips.
My setup: Ubuntu22.04;12 threads; 16GB RAM; Seismic Unix
Data Description
Water depth about 100 m
number of samples in raw data: 1950(I used only 3900ms; about 4.5km)
sample intervals: 2 ms
each shot contains 230 receivers
In FWI I use 300 shots.
src spacing: 37.5m
The processing:
f=1,5,70,80
To compute FWI I used the code:
And here is my gradient:
![Screenshot from 2023-03-17 12-29-28](https://user-images.githubusercontent.com/84370046/225814509-6d06dbc5-4651-4d1e-b9cd-2c80bd75ee76.png)
The text was updated successfully, but these errors were encountered: