Subsurface CO2 monitoring with explainable leakage detection

What is it about?

We propose an inversion-based methodology to jointly produce time-lapse seismic images without insisting on replication of surveys or labor-intensive 4D processing. Through thousands of synthetic but realistic experiments, we demonstrate that our framework is capable of producing high-resolution time-lapse images with acceptable NRMS values. We further show that a trained deep neural classifier can detect CO2 leakage in the images with explainable class activation mappings that highlight potential leakage areas in the image. We consider this work as a first step in the development of an automatic workflow to handle the monitoring of large number of CO2 injection sites in order to help combat climate change.

Featured Image

Why is it important?

We reduce the cost of time-lapse monitoring by relaxing the requirement of replicating surveys and averting labor-intensive 4D processing. We further equip our workflow with a deep neural classifier to automatically detect CO2 leakage with explainable class activation mappings to visualize the potential leakage area.