From Cloud to Clarity: Basement Fault Mapping in CO2 Storage Using a Machine Learning-Assisted Workflow

ABSTRACT: The Illinois Basin Decatur Project (IBDP) pioneered large-scale CO2 injection into a deep saline reservoir. During the project, many microseismic events were detected in the basement. It was inferred that the pressure plume reached unknown basement faults, altering their stress conditions and triggering microseismic activity. Understanding the presence and behavior of these basement faults is essential for managing and monitoring induced seismicity risks. However, basement faults with near-vertical dips are difficult to detect using conventional seismic imaging techniques due to unfavorable incident angles. In a previous study, we designed an LSTM model to map basement faults using microseismic clouds, which exhibited significant limitations, with errors ranging from 50% to 100%. To improve the model, we generated new synthetic microseismic data using a fault-zone model, which provided better correlation between the locations of the microseismic events and the fault plane. We switched to a PointNet model, a simple yet powerful deep learning network that directly consumes point clouds. PointNet, originally designed for object classification, part segmentation, and scene semantic parsing, was modified to perform a mapping task. Instead of outputting labels for each point, the modified model outputs arrays containing the probability of fault existence and fault param-eters, including location, size, and orientation. The input is solely the microseismic point cloud. The new model demonstrated a substantial improvement over the previous study, achieving an error of around 10% for fault parameters, including location, size, and orientation. This advancement underscores the potential of ML-based methods for accurately and efficiently mapping previ-ously invisible basement faults, providing critical information to enable rapid decision-making during CO2 injection to prevent and mitigate induced seismic risks. Our work showcases a fast and robust model for mapping basement faults, highlighting the significant progress made in the application of deep learning techniques to geophysical data interpretation and the management of induced seismicity in carbon storage projects. This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under contract DE-AC52-07NA27344. Release Number: LLNL-CONF-2002849