Skip to content

Python pipeline for Ambient Noise Adjoint Tomography based on SPECFEM3D_Cartesian

Notifications You must be signed in to change notification settings

enpoloo/Ambient-Noise-Adjoint-Tomography

 
 

Repository files navigation

Welcome to the Ambient Noise Adjoint Tomography python pipeline based on SPECFEM3D_Cartesian(https://geodynamics.org/resources/specfem3dcartesian).

This user-friendly package allows for easy implementation of Ambient Noise Adjoint Tomography and includes modules such as ambient noise raw data processing (ANDP-2.3), initial model, forward computation, preprocessing, measurement, adjoint simulation, pre-conditioner, kernel smoothing, and model updating. The source code of SPECFEM3D_Cartesian is provided in the 'specfem3d' folder, while the 'ANAT' directory contains 01.py-18.py, which is the workflow for Ambient Noise Adjoint Tomography. It is highly recommended to first run 'fwd_test' to determine necessary hyperparameters. 'parameter_files' stores these hyperparameters. Additionally, users will find some useful scripts in the 'scripts' directory.

If you use this python package for your research, kindly cite the following articles:
1)Zhao, Y., Guo, Z., Wang, K., & Yang, Y. J. (2021). A Large Magma Reservoir Beneath the Tengchong Volcano Revealed by Ambient Noise Adjoint Tomography. Journal of Geophysical Research: Solid Earth, 126(7), e2021JB022116.
2)Wang, K., Yang, Y., Basini, P., Tong, P., Tape, C., & Liu, Q. (2018). Refined crustal and uppermost mantle structure of southern California by ambient noise adjoint tomography. Geophysical Journal International, 215(2), 844-863.
3)Fan, X., Guo, Z., Zhao, Y., & Chen, Q. F. (2022). Crust and uppermost mantle magma plumbing system beneath Changbaishan intraplate volcano, China/North Korea, revealed by ambient noise adjoint tomography. Geophysical Research Letters, e2022GL098308.

If you have any questions, please do not hesitate to contact zhaoyanginsane@foxmail.com.

Hope you enjoy using it.

                                                   ANAT Framework

ANAT_Framework

About

Python pipeline for Ambient Noise Adjoint Tomography based on SPECFEM3D_Cartesian

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • PostScript 32.3%
  • Fortran 28.4%
  • C 20.0%
  • Perl 5.0%
  • TeX 3.1%
  • GLSL 2.3%
  • Other 8.9%