In the past few decades, artificial intelligence (AI) tools have been extensively employed in various environmental modeling tasks, including reaction modeling, groundwater level prediction, groundwater storage forecasting, and contaminant transport, to name a few. These deep learning models have been mostly limited to solving problems with large amounts of observational data because of the black-box nature of the AI tools and their inability to interpret the underlying processes. Moreover, these models cannot accurately predict the space-time distribution of transient transport processes due to the intricate and dynamic nature of transient environmental problems such as the groundwater transport problem. A new type of AI model, namely physics-infused neural networks (PINNs), offers a promising approach to solving complex, transient transport problems due to their ability to integrate physics laws into the neural network. PINNs integrate automatic differentiation methods into neural networks for directly solving the governing partial differential equation (PDE) at any point and time without a grid-based discretization procedure. In this study, we have formulated a novel custom loss function that minimizes the sum of losses from the residuals of the PDE, IC, and BC to solve a variety of governing equations. We use this approach to solve different types of transport problems using a PINN model implemented in TensorFlow. We employ various initial conditions (IC) and boundary conditions (BC) to test the performance of the PINN solution procedure for solving different types of systems. We compare the PINN solutions with the solutions derived from both analytical and numerical methods to study the advantages of PINN methods over the other solution procedures.
