About NoisePy

NoisePy is a Python package designed for fast and easy computation of ambient noise cross-correlation functions. It provides additional functionality for noise monitoring and surface wave dispersion analysis.

Disclaimer: this code should not be used "as-is" and not run like a blackbox. The user is expected to change local paths and parameters. Submit an issue to github with information such as the scripts+error messages to debug.

Detailed documentation can be found at https://noisepy.readthedocs.io/en/latest/

Citation:

Please cite the following reference if you use the code for your publication: Jiang, C. and Denolle, M. "NoisePy: a new high-performance python tool for seismic ambient noise seismology." Seismological Research Letter 91 (3): 1853–1866.

Major updates include

• adding options for several stacking methods such as nth-root, robust-stacking, auto-covariance and selective in S2. A script is added to the folder of application_modules to cross-compare the effects of different stacking method (note that substack parameter in S2 has to be True in order to use it)
• adding a jupter notebook for tutorials on performing seismic monitoring analysis using NoisePy
• adding a jupter notebook for generating response spectrum for a nodal array (to be done)

Installation

The nature of NoisePy being composed of python scripts allows flexiable package installation, which is essentially to build dependented libraries the scripts and related functions live upon. We recommand to use conda and pip to install the library due to their convinence. Below are command lines we have tested that would create a python environment to run NoisePy. Note that the test is performed on macOS Mojave (10.14.5), so it could be slightly different for other OS.

Note the order of the command lines below matters

With Conda:

conda create -n noisepy -c conda-forge python=3.7 numpy=1.16.2 numba pandas pycwt jupyter mpi4py=3.0.1 obspy=1.1 pyasdf
conda activate noisepy
git clone https://github.com/mdenolle/NoisePy.git

With virtual environment:

python -m venv noisepy
source noisepy/bin/activate
pip install wheel h5py numpy numba pandas pycwt jupyter mpi4py pyasdf
git clone https://github.com/mdenolle/NoisePy.git

To run the code on a single core, open the terminal and activate the noisepy environment before run following command. To run on institutional clusters, see installation notes for individual packages on the module list of the cluster. Examples of installation on Frontera are below.

Installing on institutional clusters:

Here is an example on how to install it on Frontera. hyp5 is already installed under the phdf5 default module.

python -m venv noisepy
source noisepy/bin/activate
pip install wheel  numpy numba pandas pycwt jupyter mpi4py pyasdf
git clone https://github.com/mdenolle/NoisePy.git

Functionality

• download continous noise data based on obspy's core functions of get_station and get_waveforms
• save seismic data in ASDF format, which convinently assembles meta, wavefrom and auxililary data into one single file (Turtorials on reading/writing ASDF files)
• offers high flexibility to handle messy SAC/miniSEED data stored on your local machine and convert them into ASDF format data that could easily be pluged into NoisePy
• performs fast and easy cross-correlation with functionality to run in parallel through MPI
• includes a series of monitoring functions to measure dv/v on the resulted cross-correlation functions using some recently developed new methods (see our papers for more details^**)

Short tutorial

0A. Downloading seismic noise data by using S0A_download_ASDF_MPI.py

This script (located in the directory of src) and its existing parameters allows to download all available broadband CI stations (BH?) located in a certain region and operated during 1/Jul/2016-2/Jul/2016 through the SCEC data center.

In the script, short summary is provided for all input parameters that can be changed according to the user's needs. In the current form of the script, we set inc_hours=24 to download day-long continous noise data as well as the meta info and store them into a single ASDF file. To increase the signal-to-noise (SNR) of the final cross-correlation functions (see Seats et al.,2012 for more details), we break the day-long sequence into smaller segments, each of cc_len (s) long with some overlapping defined by step. You may wanto to set flag to be True if intermediate outputs/operational time is preferred during the downloading process.

python S0A_download_ASDF_MPI.py

If you want to use multiple cores (e.g, 4), run the script with the following command using mpi4py.

mpirun -n 4 python S0A_download_ASDF_MPI.py

The outputted files from S0A include ASDF files containing daily-long (24h) continous noise data, a parameter file recording all used parameters in the script of S0A and a CSV file of all station information (more details on reading the ASDF files with downloaded data can be found in docs/src/ASDF.md). The continous waveforms data stored in the ASDF file can be displayed using the plotting modules named as plotting_modules in the directory of src as shown below.

import plotting_modules #(cd to your source file directory first before loading this module)
sfile = '/Users/chengxin/Documents/SCAL/RAW_DATA/2016_07_01_00_00_00T2016_07_02_00_00_00.h5'
plotting_modules.plot_waveform(sfile,'CI','BLC',0.01,0.4)                                                          

Note that the script also offers the option to download data from an existing station list in a format same to the outputed CSV file. In this case, down_list should be set to True at L53. In reality, the downloading speed is dependent on many factors such as the original sampling rate of targeted data, the networks, the data center where it is hosted and the general structure you want to store on your machine etc. We tested a bunch of the parameters to evaluate their performance and the readers are referred to our paper for more details (Jiang et al., 2020).

0B. DEAL with local SAC/miniseed files using S0B_to_ASDF.py

If you want to use the NoisePy to handel local data in SAC/miniseed format stored on your own disk, this is the script you need. Most of the variables are the same as those for S0A and thus should be pretty straighforward to follow and change. In this script, it preprocesses the data by merging, detrending, demeaning, downsampling and then trimming before saving them into ASDF format for later NoisePy processing. In particular, we expect the script to deal with very messydata, by which we mean that, seismic data is broken into small pieces and of messy time info such as overlapping time. REMEMBER to set messydata at L62 to True when you have messy data! (Tutorial on removing instrument response)

1. Perform cross correlations using S1_fft_cc_MPI.py\

This is the core script of NoisePy, which performs Fourier transform to all noise data first and loads them into the memory before they are further cross-correlated. This means that we are performing cross-correlation in the frequency domain. In the script, we provide several options to calculate the cross correlation, including raw, coherency and deconv (see our paper^* for detailed definition). We choose coherency as an example here. After running the script, it will create a new folder named CCF, in which new ASDF files containing all cross-correlation functions between different station pairs are located. It also creates a parameter file of fft_cc_data.txt that records all useful parameters used in this script. Once you get the cross-correlation file, you can show the daily temporal variation between all station-pair by calling plot_substack_cc function in plotting_modules as follows. NOTE that to make this plot, the parameter of substack has to be set to True in S1.

import plotting_modules
sfile = '/Users/chengxin/Documents/SCAL/CCF/2016_07_01_00_00_00T2016_07_02_00_00_00.h5'
plotting_modules.plot_substack_cc(sfile,0.1,0.2,200,True,'/Users/chengxin/Documents/SCAL/CCF/figures')     

2. Do stacking with S2_stacking.py\

This script is used to assemble and/or stack all cross-correlation functions computed for the staion pairs in S1 and save them into ASDF files for future analysis (e.g., temporal variation and/or dispersion extraction). In particular, there are two options for the stacking process, including linear and phase weighted stacking (pws). In general, the pws produces waveforms with high SNR, and the snapshot below shows the waveform comparison from the two stacking methods. We use the folloing commend lines to make the move-out plot.

import plotting_modules,glob
sfiles = glob.glob('/Users/chengxin/Documents/SCAL/STACK/*/*.h5')
plotting_modules.plot_all_moveout(sfiles,'Allstack_linear'0.1,0.2,'ZZ',1,300,True,'/Users/chengxin/Documents/SCAL/STACK') #(move-out for linear stacking)
plotting_modules.plot_all_moveout(sfiles,'Allstack_pws'0.1,0.2,'ZZ',1,300,True,'/Users/chengxin/Documents/SCAL/STACK')    #(move-out for pws)

Anyway, here just presents one simple example of how NoisePy might work! We strongly encourage you to download the NoisePy package and play it on your own! If you have any comments and/or suggestions during running the codes, please do not hesitate to contact us through email or open an issue in this github page!

Chengxin Jiang ([email protected])
Marine Denolle ([email protected]).

Reference

Seats, K. J., Jesse F. L., and German A. P. "Improved ambient noise correlation functions using Welch′ s method." Geophysical Journal International 188, no. 2 (2012): 513-523.
*Jiang, C. and Denolle, M. "NoisePy: a new high-performance python tool for seismic ambient noise seismology." Seismological Research Letter 91, no. 3 (2020): 1853–1866..
** Yuan, C., Bryan, J. T., Jiang, C., Okubo, K., Clements, T. and Denolle, M. (2019). Comparing approaches to measuring time- and frequency-dependent seismic phase variations for coda wave interferometry. In AGU Fall Meeting 2019.