DisbaTomo

Inversion of multimode dispersion curves of surface wave

The forwarding program is from disba, which is a computationally efficient Python library for the modeling of surface wave dispersion that implements a subset of codes from Computer Programs in Seismology (CPS) in Python compiled just-in-time with numba.

The optimization part is based on python pacakge scipy.optimize, where the gradient is implemented in C++[1], and converted into the library that can be imported by python with the help of pybind11.

Installation

There are two circumstances: with sudo permission and without sudo permission. The prior is typically for PC, and the latter is usually for clusters.

With sudo permission

Install git, eigen3 and cmake through the package manager according to the system, e.g. apt-get, homebrew, pacman.
Make sure your c++ compiler support c++14

Install python packages

Install miniconda

> wget https://mirrors.tuna.tsinghua.edu.cn/anaconda/miniconda/Miniconda3-py38_4.9.2-Linux-x86_64.sh
> bash ./Miniconda3-py38_4.9.2-Linux-x86_64.sh
> source .bashrc

Create environment

> conda create -n disba python=3.8
> conda activate disba
> conda install numpy scipy matplotlib pyyaml mpi4py pandas tqdm
> pip install disba mpi_master_slave

Without sudo permission

We use spack to install required softwares and libraries, which is a package manager for supercomputers, Linux, and macOS.

Install spack

> git clone https://github.com/spack/spack.git
> echo ". spack/share/spack/setup-env.sh" >> .bashrc
> source ~/.bashrc

Install gcc of higher version
```
> spack info gcc
> spack install ${A_VERSION_OF_GCC}
> spack location -i gcc
# spack compiler find ${LOCATION_OF_GCC_NEW_VERSION}    
> spack compiler find $(spack location -i gcc)
```
${A_VERSION_OF_GCC} is a version of gcc which you can choose from the list that "spack info gcc" shows, e.g. gcc@9.3.0.

${LOCATION_OF_GCC_NEW_VERSION} is seen from "spack location -i gcc", e.g. /home/test/spack/opt/spack/linux-centos7-haswell/gcc-4.8.5/gcc-9.3.0-swta24mlro2t4z7rfka55y3aum2ckm32

Install eigen3, git and cmake

> spack install eigen
> spack install cmake
> spack install git

Manage and load package using lmod

> spack install lmod
> echo ". $(spack location -i lmod)/lmod/lmod/init/bash" >> .bashrc
> source .bashrc
> module avail

Load package and software
```
> module load ${MODULE_NAME}
```
${MODULE_NAME} can be found from "module avail", and we need to load gcc, cmake, git and eigen3. when I installed, the command was below:
```
> module load gcc-9.3.0-gcc-4.8.5-swta24m
> module load cmake-3.18.4-gcc-9.3.0-nesdphq
> module load git-2.29.0-gcc-9.3.0-y4b63yo
> module load eigen-3.3.8-gcc-9.3.0-txefvoy
```
You can add these module load command to your .bashrc.
Install python package (the same procedure as above)

Install problems

tar (child): lbzip2: cannot exec: no such file or directory

Solution:

> spack install bzip2
> export PATH=$(spack location -i bzip2):$PATH

Error: COmmandNotFoundError: spack requires 'patch'

Solutiin:

> spack install patch
> export PATH=$(spack location -i patch):$PATH

Compilation

After installing requirements, the program is installed in the following.

> git clone https://github.com/pan3rock/DisbaTomo.git
> cd DisbaTomo
> git submodule init
> git submodule update
> mkdir build
> cd build
> cmake ..
> make -j4

Usage

Determine L-curve

> cd .. # in {DisbaTomo/}
> mkdir examples
> cp -r demo/* examples/
> cd examples/regular-large
> mpirun -np 4 ../../python/determine_lcurve.py
> ../../python/plot_lcurve.py 0

Inversion

> mpirun -np 4 ../../python/inversion.py --num_init 20
> ../../python/plot_inversion.py --data 0 --plot_model --plot_init
> ../../python/plot_inversion.py --data 0 --plot_model
> ../../python/plot_inversion.py --data 0 --plot_disp

Plot sensitivity kernel

The sensitivity is defined in [1].

> ../../python/forward.py --data data/0.txt
> ../../python/plot_disp.py disp.txt
> ../../python/plot_kernel.py model_output.txt --disp disp.txt --fmin 0.01 --fmax 0.4 --cmin 2.8 --cmax 4.5 --plot_dispersion --vmax 0.35 --mode 0 --out k0.png
> ../../python/plot_kernel.py model_output.txt --disp disp.txt --fmin 0.01 --fmax 0.4 --cmin 2.8 --cmax 4.5 --plot_dispersion --vmax 0.35 --mode 1 --out k1.png
> ../../python/plot_kernel.py model_output.txt --disp disp.txt --fmin 0.01 --fmax 0.4 --cmin 2.8 --cmax 4.5 --plot_dispersion --vmax 0.35 --mode 2 --out k2.png
> ../../python/plot_kernel.py model_output.txt --disp disp.txt --fmin 0.01 --fmax 0.4 --cmin 2.8 --cmax 4.5 --plot_dispersion --vmax 0.35 --mode 3 --out k3.png

Compute and display dispersion curves

> ../../python/forward.py -c config_forward2.yml --mode 3
> ../../python/plot_model.py model_data.txt --out model_data.png
> ../../python/plot_disp.py disp.txt --out disp_data.png

Output models when number of data is large

> ../../python/output_model.py

The script will output estimated models to the directory model_inv, which name is corresponding to the filename of data. The estimated models are based on inverted models from different initial mdoesls and their corresponding fitness value.

Configure file

TODO: translate into English

Brocher and Gardner empirical relation have been provided, and you can add other formula in python/empirical_relation.py.

References

[1] Pan, L., Chen, X., Wang, J., Yang, Z., & Zhang, D. (2019). Sensitivity analysis of dispersion curves of Rayleigh waves with fundamental and higher modes. Geophysical Journal International, 216(2), 1276–1303. https://doi.org/10.1093/gji/ggy479

ShihaoCui/DisbaTomo