Collaborative Research: CAIG: Graph Neural Network-based Digital Twins for Subsurface Geoscience

Understanding how fluids move through geological formations is hindered by complex geology and limited measurements. This project will develop artificial intelligence (AI) methods that combine observations and models to better understand subsurface systems. The project will provide student and researcher training at the intersection of AI and the geosciences. AI models, software, and benchmarks will be openly shared. These tools will support industry applications including management of subsurface geoscience. This project will develop digital twins for intermediate-scale multiphase flow experiments via graph neural networks and diffusion models. The research will create efficient surrogate models that capture multiphase flow in complex geological formations. Experimental observations will be integrated to quantify uncertainty and improve predictions. Transfer learning will adapt the models to new operating conditions and geological settings with limited additional data. These project activities will establish a scalable framework for uncertainty-aware prediction, data assimilation, and scenario exploration. The developments will be demonstrated and validated using physical experiments. Resulting methods will be broadly applicable to groundwater contaminant transport, geothermal energy, and related subsurface geoscience problems. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.