search

djpugh / MTfit

MTfit code for Bayesian Moment Tensor Fitting
78 stars 17 forks source link

Composite mechanism? #49

Open shicks-seismo opened 6 years ago

shicks-seismo commented 6 years ago

Hi again David,

I'm looking at a seismic warm with near-repeating waveforms. Polarity / amplitude data are sparse, so I'd like to compute a single moment tensor / focal mechanism for all events using all available polarities.

I really like using the NLLoc take-off angle uncertainties to compute the range of focal mechanisms, but I was wondering how easy it might be to do this for multiple events? I.e. to read multiple NLL .hyp files to compute a single composite moment tensor? Do you have any thoughts on whether you think this might be feasible? Thanks!

djpugh commented 6 years ago

Hi @shicks-seismo - short answer - yes, absolutely possible (it's how i dealt with velocity model uncertainty by sampling from multiple nlloc scatter files - see https://github.com/djpugh/MTfit/blob/develop/scripts/model_sampling.py

Without worrying about the location uncertainty to start with, you can create a composite "event" from across the data you have if a) you are happy that the events are co-located, and b) you are happy that the mechanisms are the same - MTfit has I think been used by someone else for this before, but I don't have any of the scripts for it...

There is also the relative amplitude approach, which will work for co-located events, but that doesn't work with the location uncertainty very well (marginalising the scale factor across the samples isn't implemented - I have done the maths in my thesis, but haven't implemented in code)

shicks-seismo commented 6 years ago

Thanks. I was actually ideally want to do a composite mechanism that accounts for different uncertainities for different events, but I don’t think that looks possible as it is only possible to provide a single probablity scatter file. I am wondering instead whether I can compute the MTs for each event, and then somehow “stack” the different solutions to make a single composite focal mechanism that shows all the different solutions?

From: David J Pugh notifications@github.com Reply-To: djpugh/MTfit reply@reply.github.com Date: Wednesday, 29 August 2018 at 18:20 To: djpugh/MTfit MTfit@noreply.github.com Cc: "Hicks S.P." s.hicks@soton.ac.uk, Mention mention@noreply.github.com Subject: Re: [djpugh/MTfit] Composite mechanism? (#49)

Hi @shicks-seismohttps://github.com/shicks-seismo - short answer - yes, absolutely possible (it's how i dealt with velocity model uncertainty by sampling from multiple nlloc scatter files - see https://github.com/djpugh/MTfit/blob/develop/scripts/model_sampling.py

Without worrying about the location uncertainty to start with, you can create a composite "event" from across the data you have if a) you are happy that the events are co-located, and b) you are happy that the mechanisms are the same - MTfit has I think been used by someone else for this before, but I don't have any of the scripts for it...

There is also the relative amplitude approach, which will work for co-located events, but that doesn't work with the location uncertainty very well (marginalising the scale factor across the samples isn't implemented - I have done the maths in my thesis, but haven't implemented in code)

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://github.com/djpugh/MTfit/issues/49#issuecomment-417034376, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AYmd8fPl9Uk1Ut9iseTDbGwVnTsrF5Vpks5uVs1SgaJpZM4WR744.