The new forums will be named Coin Return (based on the most recent vote)! You can check on the status and timeline of the transition to the new forums here.
The Guiding Principles and New Rules document is now in effect.
Geoscientists: What the hell is going on with this inversion
joshofalltrades
Class TraitorSmoke-filled roomRegistered User regular
Class TraitorSmoke-filled roomRegistered User regular
Hello, scientists and mathematicians.
Lately we have discovered a strange phenomenon in our data inversions. This phenomenon makes interpretation and subsequent recommendations more difficult than it should be.
I present to you a synthetic model I created.
Sorry it's so scrunched up there, but you get the general idea.
You see that double layer of light blue underneath the upper layers? The one that doesn't extend all the way to the edges of the synthetic model? That's meant to represent an aquifer.
Now, take a look at the inversion (spoilered for possible h-scroll rape):
As you can see, the color scale is changed some. I'm colorblind, but I tried to get a pinkish color for the aquifer. You can see how it extends completely to the edges of the inversion, where before it was merely in the center.
Also, and this isn't the best example of this phenomenon, but layers tend to dip at the edges of the data. This isn't entirely unexpected because of the way our electrodes work; they don't gather data straight down from their location, but rely on electrodes further along the line and the resulting data are shaped like an upside-down triangle.
My self-appointed task for the last week has been to find some way of compensating for these unfortunate tendencies that inversion has. One thing I tried (without success) is a derivative procedure:
1) first derivative @ -90 degree angle
2) make a scale recovery grid
3) multiply derivative by scale recovery grid
4) multiply derivative_recovered grid by 5
5) add the new grid to the inversion
The hope here was that this procedure would at least even out the unrealistic dips in the layers, and it did to an extent, but it also makes some layers even more unrealistic. I've tried derivatives, residuals and just plain fooling around with inversion settings, but nothing seems to produce a balanced, semi-realistic inversion based on the above synthetic model.
Am I missing something? Is there something else I should be trying? We've had successes based on our inversions before, and other times the geological conditions produce these weirdly realistic results that we'll suggest a drilling site on and it'll come up dry.
Lately we have discovered a strange phenomenon in our data inversions. This phenomenon makes interpretation and subsequent recommendations more difficult than it should be.
I present to you a synthetic model I created.
Sorry it's so scrunched up there, but you get the general idea.
You see that double layer of light blue underneath the upper layers? The one that doesn't extend all the way to the edges of the synthetic model? That's meant to represent an aquifer.
Now, take a look at the inversion (spoilered for possible h-scroll rape):
As you can see, the color scale is changed some. I'm colorblind, but I tried to get a pinkish color for the aquifer. You can see how it extends completely to the edges of the inversion, where before it was merely in the center.
Also, and this isn't the best example of this phenomenon, but layers tend to dip at the edges of the data. This isn't entirely unexpected because of the way our electrodes work; they don't gather data straight down from their location, but rely on electrodes further along the line and the resulting data are shaped like an upside-down triangle.
My self-appointed task for the last week has been to find some way of compensating for these unfortunate tendencies that inversion has. One thing I tried (without success) is a derivative procedure:
1) first derivative @ -90 degree angle
2) make a scale recovery grid
3) multiply derivative by scale recovery grid
4) multiply derivative_recovered grid by 5
5) add the new grid to the inversion
The hope here was that this procedure would at least even out the unrealistic dips in the layers, and it did to an extent, but it also makes some layers even more unrealistic. I've tried derivatives, residuals and just plain fooling around with inversion settings, but nothing seems to produce a balanced, semi-realistic inversion based on the above synthetic model.
Am I missing something? Is there something else I should be trying? We've had successes based on our inversions before, and other times the geological conditions produce these weirdly realistic results that we'll suggest a drilling site on and it'll come up dry.
joshofalltrades on
0
Posts
The first eliminates the problem of 'rough' contours caused by insufficiently many points, and the second allows me to see if I've missed some interesting feature that happens to be lying between two contours. The two don't really play well together, so it's worth trying one, then the other, then both, etc.
Apologies if this is completely useless advice.
The inversion process here consists of a program taking the raw data (which is created from the broad strokes of ohm-m levels on the synthetic), creating a forward model based on the raw data and attempting to reconcile the two.
Whilst inverting, the program keeps a histogram and provides an RMS error percentage (in the case of this synthetic model, it is only 4.32%, which is extremely reasonable). Also, the synthetic model has a small amount of noise, only 1% added to the total model. I find it hard to believe this would cause such a huge change as the aquifer extending to the edges of the data.
And no, your advice isn't useless, it's just something for me to think about.