Derisking geological carbon storage from high-resolution time-lapse seismic to explainable leakage detection

Code to reproduce results in Ziyi Yin, Huseyin Tuna Erdinc, Abhinav Prakash Gahlot, Mathias Louboutin, Felix J. Herrmann, "Derisking geological carbon storage from high-resolution time-lapse seismic to explainable leakage detection". In The Leading Edge in January 2023. DOI: 10.1190/tle42010069.1

Installation

First, install Julia, Python and MiniConda. Next, run the command below to install the required packages.

julia -e 'Pkg.add("DrWatson.jl")'
julia --project -e 'using Pkg; Pkg.instantiate()'
conda env create -f environment.yml
source activate gcs-cam
python -m ipykernel install --user --name gcs-cam --display-name "Python (gcs-cam)"

Script descriptions

We use the open-source software JUDI.jl for seismic modeling and imaging, which calls the highly optimized propagators of Devito. We used FwiFlow.jl to solve the two-phase flow equations for both the pressure and concentration. The CO2 plume dataset (consisting of regular plumes and leaking plumes) will be downloaded upon running your first example. We used PyTorch library for CAM methods to calculate the CAM images. We thank the authors of these packages for their contributions to the open-source software community.

time-lapse seismic modeling and imaging

GenLinData.jl: script to generate time-lapse linearized data via Born modeling operators.

RTM.jl: script to run reverse-time migration (RTM) on the linearized data.

JRM.jl: script to invert the time-lapse linearized data via joint recovery model (JRM).

The experimental setup (number of sources, receivers, amount of noise etc) can be adjusted according to input keywords.

To generate a dataset for training the deep neural classifier, we provide the clusterless version of the above 3 scripts --- where you can simply run the julia scripts locally and experiments can run on multiple instances in parallel on the cloud. This needs 3 files for registry, credential, and parameter information to be stored in registryinfo.json, credentials.json, params.json files. More information can be found in AzureClusterlessHPC.jl and JUDI4Cloud.jl.

leakage detection with deep neural classifier and class activation mapping

To train the deep neural classifier for leakage detection, open main.ipynb notebook and choose gcs-cam environment as the kernel. It internally uses train.py and test.py modules for training and testing. The notebook contains useful comments for each section.

LICENSE

The software used in this repository can be modified and redistributed according to MIT license.

Disclosure

Some comments in the scripts are generated by ChatGPT 4.

Reference

If you use our software for your research, we appreciate it if you cite us following the bibtex in CITATION.bib.

Authors

This repository is written by Ziyi Yin, Huseyin Tuna Erdinc, Abhinav Prakash Gahlot from the Seismic Laboratory for Imaging and Modeling (SLIM) at the Georgia Institute of Technology.

If you have any question, we welcome your contributions to our software by opening issue or pull request.

SLIM Group @ Georgia Institute of Technology, https://slim.gatech.edu.
SLIM public GitHub account, https://github.com/slimgroup.