Branched average b-values for NSHM23 WUS inversion

[heathercrume-moodys]

Can you add the weighted branch average for b-values to the NSHM23 WUS inversion so it's possible to run a single inversion with branched averaged ingredients? Thanks!

[kevinmilner]

This option is supported by the fst_inversion_factory_runner.sh script, does that work for you? It doesn't allow you to specify all of your own constraints, but rather uses the full NSHM23 inversion recipe (potentially with your data instead of NSHM23 data if you pass in a rupture set). For example, you could reproduce an NSHM23 inversion with the average b-value branch via:

fst_inversion_factory_runner.sh --nshm23 --branch-choice SupraB:AvgSupraB --output-file solution.zip

You can also pass in your own rupture set via --rupture-set /path/to/rupture_set.zip instead of building one for NSHM23 data.

It will be a bit trickier to build this into the regular and more customizable inversion tool. Instead, I suggest averaging them post hoc, which is possible with the solution averager tool (and simple because they're evenly weighted).

[heathercrume-moodys]

This works great, thank you! One follow-up question, are the averages in the different branch options (for example, --branch-choice DM:AVERAGE) the weighted averages from the logic tree or are they equally weighted?

[kevinmilner]

Good question; they're the weighted average from the model. So, uneven for deformation model and segmentation, even for b-values.

If you use the average b-value and/or segmentation branches, it does a pretty good job of approximating the average of separately-run inversions because it calculates the target MFDs separately for each branch choice, then weight-averages those target MFDs. This gets better to the average behavior than using, for example, a central b-value or an segmentation penalty.

[heathercrume-moodys]

Perfect, thank you!!