mewest commented 1 week ago

Spike detection appears in multiple LTS classes. I can't sleuth out how the term spike is being defined algorithmically? Is this a common issue in infrasound data? I don't see it discussed in Jordan's paper.

Cade and I are feeding pretty clean seismic waveforms into ltsva and getting a large percentage of nan's out the other end. Spike detection may or may be the cause, but it caught my attention. The waveforms produce fine results out of obspy/array_processing, but we are interested in trying a least squares solver.

https://github.com/uafgeotools/lts_array/blob/5a52703a628e5c12d3945daf88dcad1b3b9f87f8/lts_array/classes/lts_classes.py#L910C1-L913C25

davidfee5 commented 1 week ago

I think this was related to situations where "spikes" in the data occurring at the exact same time across multiple channels would produce a time delay of zero. I'd be surprised if this was your issue and imagine you could easily check by just commenting it out and trying again. I'd be happy to look at some example waveforms/etc to help you diagnose why are you getting nans.

caquigley commented 2 days ago

Attached is a summary document of the issues we've been having and the temporary fix we've been using for our data. The fix is related to removing some information in the spike detection part of the code. Below is the user inputs for the example.py for our dataset of interest. Let us know what you think.

Issues with lts_array spike detection.pdf `# User Inputs

Filter limits [Hz]

FREQ_MIN = 2 FREQ_MAX = 4.0

Window length [sec]

WINDOW_LENGTH = 2

Window overlap decimal [0.0, 1.0)

WINDOW_OVERLAP = 0.5

LTS alpha parameter - subset size

ALPHA = 1 #Look at least squares example

Plot array coordinates

PLOT_ARRAY_COORDINATES = True

#####################################################

End User inputs