NSF Award
2425919
Subsurface flow and transport systems within the Earth experience complex interactions among rocks, fractures, and fluids. Subsurface system dynamics affect groundwater aquifers, geothermal energy, hydrocarbon resources, geologic carbon storage, and other important Earth resource management needs. A significant challenge in predicting the behavior of these subsurface systems arises from the need to specify highly uncertain rock physical properties, which are inherently heterogeneous and exhibit variability across multiple scales. Advances in artificial intelligence (AI) offer unprecedented capabilities for processing large and diverse multimodal datasets. In this project, advanced AI models will be developed to capture salient spatial and temporal relations in subsurface flow and transport systems by integrating physical principles and existing multiphysics data. This work has immediate impact on significant societal geoscience applications, such as groundwater aquifers and geothermal energy recovery, as well as climate change and environmental sustainability. The project will attract and educate students underrepresented in STEM fields and equip the next generation of the geoscience workforce with AI skills.<br/><br/>The objective of this project is to develop novel domain-aware, robust, and interpretable AI solutions for capturing and predicting subsurface flow and transport dynamics. Specifically, Physics-Informed Causal Deep Learning Models (PINCER), will be developed to detect and exploit spatial and temporal relations in subsurface flow and transport systems by integrating physical principles and multiphysics data. This research is focused on developing: (1) novel deep learning architectures that honor the general structure of fluid flow equations while accounting for uncertainties and allowing for learning and adaptability based on observed monitoring measurements; (2) physics-informed causal deep learning models for succinctly capturing and predicting fluid flow and transport dynamics; and (3) flexible deep learning-based inference of heterogeneous rock flow properties from incomplete multiphysics monitoring data. PINCER presents a paradigm shift from traditional data-driven approaches or model-based techniques to a hybrid solution that combines the benefits of both methods. It makes novel contributions in several AI research areas, such as causal analysis, physics-informed AI, and interpretable AI. It also advances geoscience research by developing more efficient and robust modeling and prediction of fluid flow and transport processes in subsurface environments. The PIs will broaden the impact of their work by training students, disseminating research findings (including new datasets and open-source software), and developing outreach programs.<br/><br/>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.
