Application of Deep Learning Models for Groundwater Data Analysis: A Comparative Study of CNN (Conv1D), SimpleRNN, and Gated Recurrent Unit (GRU) Models

Abstract

Groundwater is a crucial resource for managing water sustainably, but it faces challenges like overuse and changes due to human activities and climate. This study looks at how deep learning models—specifically Convolutional Neural Networks (Conv1D), Simple Recurrent Neural Networks (SimpleRNN), and Gated Recurrent Units (GRU), can be used to analyze groundwater data from three monitoring stations in Vietnam. The data comes from different areas, including urban (Hanoi), metropolitan (Ho Chi Minh City), and rural-agricultural (Kien Giang) regions, with varying time intervals and data characteristics. The models were tested using several performance metrics like Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-squared (R²). The results showed that the Conv1D model was the best, providing the most accurate predictions, especially when the data had moderate changes. The SimpleRNN model worked well with high-resolution data but struggled when the data was more variable or incomplete. The GRU model had limited success with data that showed significant fluctuations. This study shows that Conv1D is a strong tool for groundwater monitoring and offers useful guidance on choosing the right model based on the data. It also suggests that improving data handling and adding more relevant features could help further improve predictions.