Meetings¶

Probably we will aim to meet each week for the next few weeks until end of September.

---

3. 2023-09-12 Wednesday¶

PE, JS, FT (Sorry, I'm very late with this summary --P.)

We met fairly informally.
Previously on 11.09 I'd spoken with Dorit K. from AG Microsites. See Microsites Notes on the wiki here.

• Data access/Terms of Use. One key point from the discussion with AG Microsites: Providers of datasets/models on the repository should take care that their terms of use / licenses are settled, and clearly displayed. It would be best if (i) there could be a terms of use statement for each dataset there, and (ii) instructions to contributors make clear that such a statement is expected.

• Other work:
  - I'm looking at understanding/running the seisbench/pick-benchmark example.
  - Joachim is still working through a script to collect the data for training, and can show me when he's ready.

2. 2023-09-05 Wednesday¶

PE: Have training in 3a example working on the laptop; haven't studied the second example from Jannes. Took a few hours for ETHZ dataset (22Gb).

[Didn't mention: discussion with GFZ about the data repository mirror is moving along; I have a meeting on 11 Sept about this.]

It's not clear where inventory came from in the current geofon data set. JS is working on clearing this up for the new work. Where and how this new code should be committed (and published) isn't clear yet either. JS has a repo for "SB_stuff" exploration.
PE and JS will work on understanding this.

Working from local files instead of public FDSNWS dataselect means some data may not be openly available, or has to be removed from processing (e.g. IA).

ETHZ data set is 22GB and took "days" to build; GEOFON (26GB) will take longer.

JM: Hank @NEIC is ready with his updated dataset. This will be transported to Grenoble, then dCache (Hamburg) then us somehow.

[Gossip about which picker is best omitted.]

---

1. 2023-08-30 Wednesday¶

FT, JM, JS, PE.

Current "GFZ" dataset covers events until 2013. We would like to now use events from 2013-202x (2020?).
From around 2013 Joachim had a new interpretation of picking.

NEIC: Hank @NEIC has/will release a dataset covering 2013-2022.

How to do it? Look at data/geofon.py

Need to choose and clearly define split of the data into training and test data.

• The SCDL picker: epicentre + traces --> depth estimate. Questions for the future:
  1. How to extend this to routine use for all processed events?
  2. How to return the result back into SC on proc? At least: flag suspicious cases with large discrepancies from the existing conventional SC depth.

Organisation: Try to meet weekly.

Misc:

• Thomas Borstein (Co-author on the Woollam paper): script to convert waveforms associated to events to SeisBench format.
• Michael Fritzsch? @KIT.
• Marius Isken: LASSI v2 @GFZ.
• MLaapede?