Abstract for: Forecasting Groundwater Storage Changes: A Deep Learning Model Leveraging System Dynamics Simulation Data

Groundwater is crucial for agriculture, industry, and domestic use, necessitating effective monitoring and predictive methods. Traditional numerical models, while accurate, face operational challenges due to their dependence on extensive, hard-to-obtain data. This research introduces a new approach that integrates System Dynamics (SD) with Deep Learning (DL) to forecast groundwater storage changes. The study uses a validated SD model from the water-scarce Lower Rio Grande (LRG) region of Southern New Mexico. Outputs from the SD model inform the DL process, utilizing an Artificial Neural Network (ANN) to streamline groundwater dynamics by concentrating on a select set of variables. The DL model achieved a training RMSLE of 0.031 and an R-squared of 0.77. For testing, the RMSLE was 0.036 with an R-squared of 0.71, demonstrating predictive reliability. These results highlight the model’s utility in forecasting future groundwater changes with reduced dependency on extensive data acquisition. Notably, this research pioneers using SD simulation data to train DL models as a data augmentation method, suggesting broader applications and emphasizing system dynamics as a valuable tool for deep learning in hydrological modeling. The findings enhance groundwater modeling, supporting informed policy and management strategies.