Possibly incorrect numerical annotations in waveforms plot

[aakash10gupta]

Dear MTUQ developers,

While running moment tensor inversions for simple earthquakes, I'm noticing that the values for the numerical annotation "contribution to the misfit" (third row below each seismogram), in the waveforms plot, are all 0s for data-syn comparisons for all components at all stations. I have verified this with an MTUQ example in the examples folder.

I'm using the latest MTUQ package environment. Any insight into this will be very helpful. Thanks.

[carltape]

Hi folks, it would be great if we could resolve this, since the waveform fits plots do not provide the "contribution to the misfit" information. It lists 0.00 for every station and every component.

[ammcpherson]

@rmodrak @thurinj I believe this is a simple fix, and is due to the new misfit normalization. @aakash10gupta has identified that this is not an issue before the September 23 update, and I am using the September 23 update where it is also present.

I believe the root of this issue is here, where when plot_data_greens[1,2,3] generates the synthetics with all the attributes attached. By default, MTUQ explicitly sets normalize=False when attaching the synthetic attributes.

This specific issue can likely be solved by removing the explicit normalize=False in the code I linked, but I did not do any more digging to see if there are any more instances that need to be updated.

Thanks

[aakash10gupta]

Just to be precise, the old version that I test that correctly shows the annotation is commit 55008d4.

@ammcpherson, you may want to edit the date you mentioned to avoid any confusion.

[ammcpherson]

The change I am speaking of occurred on September 23 in commit 6b93b08

[ammcpherson]

@rmodrak @thurinj I believe I have solved the issue locally. In level0.py, when the individual trace misfit is attached to the synthetics, there is no normalization applied. Adding a if normalize: value /= norm_data before attaching the misfit as an attribute solved this issue for me. I have also changed normalize options in 3 places: 1, 2, 3.

[thurinj]

Hi @aakash10gupta, @ammcpherson, please check PR #333 , as I've addressed this issue in it. I will close when merging.

[aakash10gupta]

I tested the branch waveform_plot_update out. Looks like the annotation problem is resolved.

I did notice some unrelated issues with the new branch though -

• The clone of the branch resulted in the download of a 607MB package, instead of the 314MB for the latest master branch. git clone --branch waveform_plot_update https://github.com/mtuqorg/mtuq.git
• The log message Normalized misfit stored in trace attributes. repeats hundreds of times, probably once for each trace.

[thurinj]

Ah, thanks @aakash10gupta, I forgot to remove the print statement that I used for debugging. Good that I didn't pull the trigger right away.

I am going to check how to solve this size issue, i suspect it comes from the merge operation I did at some point.

[thurinj]

Quick follow-up > it was actually way easier to just pop a clean branch and re-create the modification from there (I am not even sure why there was a conflict when I worked on the other branch).
I've made PR #339, and will delete both branch, and old pull request once merged (before the end of the week if there are no objections).

Thanks again for catching the bug and helping troublehsoot @aakash10gupta and @ammcpherson !

[thurinj]

Solved with #339