adjtomo / pyadjoint

Measure waveform misfits and generate adjoint sources for a variety of measurement quantities

Pyadjoint

Misfit measurement and adjoint source generation

• Pyadjoint is a small misfit measurement package that calculates misfit and adjoint sources on time series
• Documentation for Pyadjoint can be found on GitHub
• Can be used standlone, but is also wrapped in Pyatoa (alongside Pyflex)
• Pyadjoint is part of the adjTomo ecosystem of tools for adjoint tomography and full waveform inversion

pyadjoint's Issues

Issue with pyadjoint during execution of Seisflows example 2

Pyadjoint ver: 0.2.1

Doc broken

Some parts of online doc is broken

adopt or merge DD and envelope misfit functions from computational-seismology

There is a separate fork of PyAdjoint which contains a significant amount of development which is not present here. From what I can see these include double difference (DD) and exponentiated phase misfit functions. I think there are also a number of bug fixes in that branch that are not present here. Although I am not aware of specifics.

https://github.com/computational-seismology/pyadjoint/tree/dev

I was thinking we should try to merge these changes into this main repository so that they're accessible to all, and usable for a SeisFlows or Pyatoa workflow.

Looking through the commit history, @chukren @yanhuay @wjlei1990 @rdno are the main contributors. Questions for you all:

• Are you using the computational-seismology version of Pyadjoint?
• Do you have any objection with attempting to merge these two repositories, or somehow co-opting those changes in this repository?
• Is there any summary of diverging changes made, besides the commit history?

A typo for writing adjoint source in ASDF format

https://github.com/krischer/pyadjoint/blob/e42f24acd46a981b85a57ca4039313e1886f1223/src/pyadjoint/adjoint_source.py#L190

Hi Lion,
It seems that this line should be changed to specfem_adj_source[:, 0] += time_offset .

Transfer this repo

Hi, Please transfer this repo to:

https://github.com/computational-seismology

Adjoint source for single component data

Hi,
Many recent seismic data are recorded by nodal arrays, and they are single component with channel code "DPZ". For the kernel simulation in SPECFEM3D, elastic simulation requires three component (X,Y,Z) adjoint source data to be read in by the code.
How can I prepare the adjoint sources for X and Y component, even if they don't exist? My thinking is that I manually create two zero traces (i.e., repeat zero by NSTEP times), input them to pyadjoint, and output them in specfem adjoint source file format. Now I have three component adjoint source data, but only Z-component is the real valid trace.
Just want to make sure I'm doing right and there's no shortcut to do this. Thanks.

sensitivity kernel (banana-doughnut) adjoint source

Include sensitivity kernels as a valid adjoint source type, which is simply feeding a manipulated form of the synthetic seismograms back as the adjoint source. This has already been coded up here:

https://github.com/adjtomo/pyatoa/blob/3690bc8bc59c76dcfdf31885259dc55a4efd681e/pyatoa/utils/adjoint.py#L9-L63

But will need to be written in PyAdjoint formatting