Applications of Physics-Informed Neural Networks in Groundwater Science: Hydraulic Parameter Estimation and Groundwater-level Monitoring Data Quality Management

This study validates the applicability and advantages of Physics-Informed Neural Networks (PINNs) in groundwater science using real-world field data.Specifically, we demonstrate PINN's effectiveness in estimating aquifer hydraulic properties and detecting outliers and handling missing values in real-time groundwater level observations.The developed PINN model incorporates a Linear Reservoir Model (LRM) as the physical constraint.Daily groundwater level and precipitation data collected from 14 monitoring wells across Jeju Island during 2010-2024 were utilized for training regional groundwater level variation patterns.Twelve years of data (2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019)(2020)(2021) were allocated for PINN model development, while observations from 2022 to 2024 were utilized for model performance validation.Aquifer hydraulic parameters were estimated through Bayesian optimization and validated against optimal parameters derived from the analytical solution of LRM, confirming the physical interpretability of the developed PINN.For outlier detection and missing data imputation, the PINN-predicted normal range of groundwater levels effectively identified anomaly caused by sensor malfunctions or environmental disturbances.These results demonstrate that the developed PINN provides an effective approach to address the lack of physical interpretability in purely data-driven models and to enhance the quality of real-time observation data.