Toward Automated Early Detection of Risks for a CO 2 Plume Containment From Permanent Seismic Monitoring Data

Permanent reservoir surveillance is an invaluable monitoring tool for CO 2 storage projects, because it tracks spatial‐temporal evolution of the injected plume. The frequent images of CO 2 plumes will facilitate history‐matching of the reservoir simulations and increase confidence of early leakage detection. However, continuous data acquisition and real‐time interpretation require a new approach to data analysis. Here, we propose a data‐driven approach to forecasting future time‐lapse seismic images based on the observed past images and test this approach on the Otway Stage 2C data. The core component of the predictor is a convolutional neural network, which considers subsequent plume maps as color layers, similar to standard red‐green‐blue blending. Based on the extent of the past plumes, we may predict the future contour of the seismically resolvable portion of the plume. The neural networks reproduce the dynamics of CO 2 migration after training on reservoir simulations for a wide range of injection scenarios and subsurface models. Extensive testing shows that realistic plumes for Stage 2C are too complicated and the neural network should be pretrained on simpler reservoir simulations that include only one or two geological features, such as: faults, spill‐points, etc. Such staged training can be seen as a gradual descent of the neural network optimization to a global minimum. The approach is practical, because each CO 2 storage project requires extensive preinjection reservoir simulations. Once the predictor has been trained, it can forecast plume evolution near real‐time and adapt efficiently to changing dynamics of CO 2 migration.