Closed OlgaPuntous closed 4 years ago
Estimating an error model (noise level) for real data is an essential step as it defines the regularization strength to be chosen and therefore the smoothness of the solution and the resolution properties. Looking at the data is therefore essential, but often you have rules of thumb from experience, by picking first arrivals or from considerations regarding frequency content and fresnel volumes. In Refraction seismics you often have an impression of the noise from how oscillating the traveltime curves are, in crosshole tomography it is a bit more complicated. What you can do is to look at reciprocal data, i.e. differences between traveltimes with interchanged shot/receiver positions, if available. At any rate it is nothing we can answer here as it is specific to your data, but any hints on estimating noise from others are welcome. In general, I can only recommend starting with a conservative guess and become more and more optimistic until artifacts show up in the model, then you're a bit too far.
Note that in the above code there is some inconsistency:
noiseLevel=0.001, noiseAbs=1e-5), where is 0.1% relative and 10 microseconds of absolute noise.
The relative noise level is only for stability and not important here.
Thanks a lot Thomas for such a quick and clear answer.
how to display data and response values in 2D ERT modelling and inversion?
how to display data and response values in 2D ERT modelling and inversion?
This question does not seem to fit here. There are examples of 2D ERT including data and response at pygimli.org
Problem description
Hello,
I am trying to make cross hole travel time tomography Pygimli like in example. In example programm invert synthetic data, produces by function simulate data
tt.simulate(mesh=mesh_fwd, scheme=scheme, slowness=1. / model, noiseLevel=0.001, noiseAbs=1e-5), where is 0.1% relative and 10 microseconds of absolute noise.
In my case I would like to invert real traveltime data. I noticed that invertion result depend a lot from noise level input.... Which noise level I should introduce in my real data to get moreless stable result?
Thanks a lot for your advice.
Your environment
Windows
Python version: 3.6 pyGIMLi version: 2019 Way of installation: e.g. Conda package
Steps to reproduce
Tell us how to reproduce this issue. Ideally, you could paste the code that produces the error:
Expected behavior
Actual behavior