Modeling of daily groundwater level using deep learning neural networks

Abstract

Groundwater is an essential water source, becoming more vital due to shortages in available surface water resources. Hence, monitoring groundwater levels can show the amount of water available to extract and use for various purposes. However, the groundwater system is naturally complex, and we need models to simulate it. Therefore, we employed a deep learning model called CNN-biLSTM neural networks for modeling grounding, and the data was obtained from USGS. The data included daily groundwater levels from 2002 to 2021, and the data was divided into 95% for training and 5% for teasing. Besides, three deep CNN-biLSTM models were employed using three different algorithms (SGDM, ADAM, and RMSprop. Also, Bayesian optimization was used to optimize parameters such as the number of biLSTM layers and the number of biLSTM units. The model's performance was based on Spearman's Rank-Order Correlation (r), and the model with SGDM showed the best results compared to other models in this study. Finally, the CNN model with LSTM can simulate time series data effectively.