Inversionson: Fully Automated Seismic Waveform Inversions

We present Inversionson, a Python package that fully automates modern full-waveform inversions (FWI).
It supports traditional FWI, which uses the same set of events and a single simulation mesh in each iteration, as well as more advanced workflows that exploit the use of dynamic mini-batches and wavefield-adapted meshes.
These recently introduced advancements can be time-consuming and challenging to implement but offer major benefits, such as much lower computing cost in favourable cases and the ability to use much larger datasets, while having a built-in optimal experimental design.
Inversionson, in essence, acts as a control center...  more

We present Inversionson, a Python package that fully automates modern full-waveform inversions (FWI).
It supports traditional FWI, which uses the same set of events and a single simulation mesh in each iteration, as well as more advanced workflows that exploit the use of dynamic mini-batches and wavefield-adapted meshes.
These recently introduced advancements can be time-consuming and challenging to implement but offer major benefits, such as much lower computing cost in favourable cases and the ability to use much larger datasets, while having a built-in optimal experimental design.
Inversionson, in essence, acts as a control center that steers the functions of several other packages, making them communicate and work together to perform seismic inversions.
This enables scaling of full-waveform inversion problems well on the most powerful High Performance Computing (HPC) clusters while requiring no user input during run time.
Widespread adoption of real data full-waveform inversion workflows to image regional to global 3-D Earth structure has so far been limited to a handful of research groups, with most researchers focusing on 2-D synthetic problems.
This probably mainly stems from the challenging implementation, as well as a lack of computational and human resources.
Inversionson is built on top of various tools and mitigates each of these challenges, by exploiting the widely used waveform solver package Salvus, supporting efficient modern workflows, and fully automating the time-consuming steps.
With this contribution, we hope that we can make FWI a more interesting option for a much larger group of researchers, thereby accelerating progress in the field.