A Reduced Order Basis Approach For Co2 Monitoring From Sparse Time Lapse Seismic Data

Download or read book A Reduced-order Basis Approach for CO2 Monitoring from Sparse Time-lapse Seismic Data written by Badr Waleed A Alrumaih and published by . This book was released on 2019 with total page pages. Available in PDF, EPUB and Kindle. Book excerpt: I present an approach for seismic monitoring from sparse time-lapse data, with a particular focus on leak detection from CO2 storage reservoirs. I use sparse data because it is (1) faster and (2) less expensive to acquire and to process, permitting for more frequent monitoring surveys to be carried out. This would allow for (1) early leak detection, which is what we ultimately aim for at a storage site, and (2) timely assessment of performance conformance. To account for data sparsity, I incorporate information on the underlying (injection) process (pressure and flow) into the geophysical model estimation. By process information, I mean how the geophysical model is possibly or potentially perturbed due to CO2 injection, as governed by the physics of the flow and the rock properties model. I do that by reformulating the geophysical minimization problem with Reduced-Order Basis (ROB) functions that are derived from simulated training images stochastically describing how the geophysical model is perturbed by the CO2 injection including leak possibilities, which I will refer to as ROB-inversion. Naturally, reducing the spatial sampling of the acquired data leads to reduced spatial resolution of the reconstructed subsurface model. This is the tradeoff for the increased calendar-time resolution, i.e., the shorter monitoring calendar-time interval. By reformulating the geophysical minimization problem with the process-derived reduced-order basis functions, I can improve the spatial resolution of the subsurface model—leading to approximate (or reduced-order) models. The accuracy of the reduced-order models depends on how representative the training image set is to the true model change. A key point in my implementation is the formulation of the problem in terms of the changes in model and data—not in terms of model and data. This (1) focuses the inversion on the model change, making it easy to apply restrictions and limitations on the model change during seismic inversion; the ROB-inversion essentially restricts the model change to be in terms of the (process-derived) Reduced-Order Basis functions. Furthermore, it (2) allows for the training images to be defined explicitly in terms of the time-lapse changes to the baseline model. The change is generally constrained—by the physics of the flow and the rock properties model, making a representative training image set to be reasonably attainable. An advantage of my approach over existing sparse time-lapse techniques is that it allows for fixed data acquisition configurations over calendar-time. Hence, the cost and turn-around time associated with redeployment of seismic data acquisition equipment can be minimized. In order to demonstrate my approach, I focus on borehole-based monitoring, namely, crosswell data acquisition geometry; nevertheless, it can be adapted to other geometries (surface-based or borehole-based) and other geophysical data (e.g., resistivity, electromagnetic, etc.). It can also be adapted for monitoring other processes, such as assessing the performance of Improved Oil Recovery (IOR). In this thesis, I demonstrate the practicability of my approach on synthetic and field traveltime crosswell datasets. I show, with synthetic and field data, its effectiveness for leak detection during CO2 injection.