Why Anxcor ? - uofuseismo/anxcor GitHub Wiki

Crosscorrelation functions derived from Ambient Seismic Noise have broad applicability across geophysics, from monitoring volcanic activity [Lobkis and Weaver, 2001], to informing seismic vulnerability assessments [Prieto, GA and Beroza GC 2008]. The advent of cheap geophones like the Magseis-Fairfield Zland 3C have research departments and USGS offices flush with data. These data range from the gigabyte to multi-terabyte scale, and can be a pain to keep organized.

Typical crosscorrelation workflows include read/write operations at every step of signal processing. Even with downsampling and Signal-Noise Ratio data curation, directories can easily become crowded with thousands of files. The problem is exacerbated by attempts to shove every bit of meaningful metadata into file names, creating a logistical nightmare for all but those intimately familiar with the datasets. Additional, the optimized code used to analyse these data often use workflows with a mish-mash of bash, tcsh, sac, fortran, and c scripts; compiled codes with succinctly (but unreadable) defined variables, and few modularized functions.

Solutions like MS Noise, are excellent, well-cited, and stable. At the risk of introducing yet another standard, we present ANXCOR as a possible solution for data scientists with limited knowledge of cluster compute workflows.

We mitigate the above problems by using xarray datastructures for vectorized computations and metadata persistence. Use of dask, NumPy, and SciPy, allows us to create an embarrassingly parallel compute graph. This allows computation to take place almost entirely in RAM, eliminating redundant file proliferation, while allowing the user to select specific outputs to save to file.

Anxcor also provides abstract classes useful for the user who would like to implement their own correlation methods or preprocessing steps. Because we defer all parallelization to be handled by dask, we can make such interfaces concise, readable, and highly modular.

Because ANXCOR is aimed specifically at python's datascience ecosystem, we believe we can provide an approach to seismic ambient noise crosscorrelation that provides great utility to users already familiar with these packages.