Physics-informed and data-driven machine-learning approaches for characterizing solute transport and risk analysis in heterogeneous aquifers

Groundwater is a key source of freshwater on which about one-third of the global population depends. Contamination of groundwater, originating from industrial and agricultural activities, has raised concerns over the potential risks to the environment and public health. The proposed work aims to develop a novel approach that will improve our fundamental understanding of contaminant transport mechanisms in the subsurface environment.

More specifically, the researchers will utilize this framework to shed novel insights into how the spatial structure of the subsurface geologic formation interacts with the large-scale spreading behavior of the contaminant plume. Such insights are fundamental to predicting the risks associated with groundwater contamination, and to define optimal contaminant remediation strategies.

These predictions can help keep groundwater resources sustainable. The proposed approach makes use of machine-learning algorithms and computer models that will enhance our ability to manage contaminated groundwater resources. Furthermore, a set of educational activities for middle school, undergraduate, and graduate students is proposed.

The researchers highlight a novel educational game that aims to show undergraduate students the role of the geologic structure in groundwater contaminant movement.

The proposed work consists of the development of a hybrid computational framework that combines the benefits of physics-based models and data-driven methods in order to enhance our understanding of the relationship between the heterogeneous structure of the geologic formation and the overall spreading of the solute plume. The approach makes use of machine learning algorithms, stochastic optimization, and physically-based models utilizing highly-efficient GPU-based simulations.

The proposed computational framework will efficiently estimate dispersion coefficients and identify the governing equation for the solute concentration field in macroscopically heterogeneous porous media for which the representative elementary volume is on the order of the system’s size. The outcome of this work will be tested against field data and will lead to great improvement in the predictive capabilities of hydrological models.

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.