Machine Learning-based 4-D Seismic Data Integration and Characterization of Channelized Anticline Aquifer for Geological Carbon Sequestration

This study introduces a 4-D seismic data integration approach based on Pix2Pix to quantify uncertainty and enhance reliability in subsurface CO2 storage within channelized aquifers. We address geological uncertainties, such as the azimuth of channel direction, using multi-point statistics to generate diverse channelized aquifer models, to form the Pix2Pix training dataset. After successfully training the Pix2Pix, we realize the channelized aquifer models by assimilating 4-D seismic data changes from a reference channelized aquifer. The Pix2Pix dataset is crafted with (1) disconnections in artificial neural networks using dropout and (2) the addition of random noise to the input layer, yielding an ensemble of models that align with the provided data for quantifying uncertainty. Generating channelized aquifers using the suggested method improves prediction of channel direction, facies distribution, and accurate forecasts of future CO2 plume movement. As an advanced monitoring technology, our method integrates 4-D seismic data for effective CO2 storage in channelized aquifers.