PINN-Based Spatio-Temporal Workflow for Groundwater Risk Prediction and Decision Support

Model Introduction and Process: Groundwater systems that sustain more than 2.5 billion people are increasingly stressed by the compound pressures of climate variability and rapid urbanisation. Traditional numerical models struggle to assimilate sparse observations and physical constraints simultaneously. We therefore develop a physics-informed neural network (PINN) that embeds the transient groundwater-flow equation directly into the loss function, achieving physically consistent hind-casts and forecasts while capturing fine-scale heterogeneity. This code implements a Physics-Informed Neural Network (PINN) for groundwater modeling. In this case, the PINN is designed to solve a groundwater flow problem by minimizing not only the error between its predictions and the observed data but also the error in satisfying the governing partial differential equation (PDE) for groundwater flow. This helps the model learn physically consistent solutions, even with limited data. This full script is ready for Colab (T4 GPU or similar) and delivers state-of-the-art accuracy for physics-informed modeling of groundwater heads. Remember to change the path "/content/groundwater_pinn_dataset.csv" in Colab if directly uploaded.