Behnam Jafarpour

University of Southern California

AI-enabled sparse subsurface monitoring: from real-time imaging to data acquisition design and optimization

Cost-effective monitoring that extracts maximum information from sparse, expensive field measurements is a major challenge, compounded by significant geologic uncertainty. This talk presents an AI-driven framework for subsurface monitoring that jointly images reservoir states in real time and adaptively optimizes data acquisition under uncertainty. Operating solely on measurable field observations, without unobservable reservoir properties or costly full-physics simulation at deployment, the framework is trained offline on large ensembles of physics-based simulations spanning diverse geologic realizations, learning uncertainty-aware relationships between observations, subsurface properties, and dynamics. By integrating deep learning, physics-based simulation, geologic uncertainty, and information-driven optimization, the developed deep learning model learns not only to interpret sparse data in real time but also where to acquire it for maximum predictive value and best allocate monitoring resources. Case studies from CCUS applications show joint imaging and monitoring optimization to reduce sampling requirements without sacrificing predictive accuracy, offering a scalable, computationally efficient path toward intelligent, uncertainty-aware monitoring for safe, large-scale subsurface deployment