Assessing the accuracy of fault interpretation using machine-learning techniques when risking faults for CO2 storage site assessment

Generating an accurate model of the subsurface for purpose assessing feasibility a CO 2 storage site is crucial. In particular, how faults are interpreted likely to influence predicted capacity and integrity reservoir; whether this through identifying high-risk areas along fault, where fluid flow across or by reactivation potential fault with increased pressure, causing up fault. New technologies allow users interpret effortlessly, in much quicker time, using methods such as deep learning (DL). These DL techniques use knowledge from neural networks end compute occur. Although these new may be attractive due reduced interpretation it important understand inherent uncertainties their ability predict geometries. Here, we compare versus manual interpretation, can see distinct differences those significant ambiguity exists poor seismic resolution at fault; observe irregularity when used over conventional interpretation. This result between resulting analyses, potential. Conversely, that well-imaged indicate close similarity surfaces used; hence, also any attributes analyses made.