This repository contains scripts to run numerical simulations of seismic monitoring of CCS with joint recovery on Azure batch. This software is based on JUDI, which uses the highly optimized time-domain finite-difference propagators of Devito and explores the serverless parallelism through JUDI4Cloud.

First, please put your Azure credential file in this directory (Seis4CCS.jl/credentials.json). e.g. do

touch credentials.json
vi credentials.json

and copy-paste your credentials to this file.

Then, under Seis4CCS.jl, run

L=a nsrc=b vm=c nth=d niter=e bs=f snr=g gamma=h bash runall.sh

where integer a is the number of vintages in the experiment (maximum is 879), integer b is number of jittered sample seismic sources in each vintage (different for each vintage, suggested to be more than 8), c is the type of virtual machine for seismic data simulation and imaging, d is the number of threads on each computing node in Azure batch, e is the number of joint recovery iterations, f is the batchsize in a JRM iteration, float g is the SNR of the noise to be added in the data, float h is the weight gamma on common component (suggested to be a/2). The default values of a,b,c,d,e,f,g,h are 2,8,Standard_E8s_v3,4,16,4,0,1.0.

When the aformentioned program finishes, the generated seismic data will be saved in ./data folder and each iteration of joint recovery results will be saved in ./results folder. To test if everything runs without error in a small scale, you could do

bash runall.sh

These scripts are suggested to run from docker image, which will be provided upon request.

Notice that if you set L, nsrc, niter to be very large, it may require massive disk space and memory on your local machine. It is also suggested to add --oom-kill-disable when you run the docker image.

There is rclone installed in the docker image. Feel free to use it to upload the results to your preferred online platforms.

If any question, please contact

Ziyi (Francis) Yin, [email protected]
Mathias Louboutin, [email protected]
Felix J. Herrmann, [email protected]

This research was carried out with the support of Georgia Research Alliance and partners of the ML4Seismic Center.