Groundwater Level Estimation using Recurrent Neural Networks: A Case Study of the Grootfontein Aquifer

Abstract

Precise estimation of groundwater levels is essential for the management and sustainability of groundwater resources. The main objectives of this study are therefore (1) to quantify the influence of groundwater extraction, precipitation, temperature and discharge on the prediction of groundwater levels in the Grootfontein aquifer, (2) predict groundwater levels under different climate and groundwater extraction conditions using a recurrent neural network and (3) compare results of each case scenario with the base case for analysis. Selected datasets from feature analysis were fed into a recurrent neural network architecture to simulate the seasonal groundwater level changes. Feature analysis results revealed that the variables selected indeed had a strong influence on the prediction of groundwater levels on the selected boreholes. Discharge, groundwater abstraction and precipitation were the highest contributing factors to groundwater level fluctuation. A recurrent neural network model was used to simulate different case scenarios. The model results reveal that the neural network model was able to predict groundwater level change under the adjusted input variables. Groundwater level fluctuations were no more than 2 m below the base case for all scenarios tested. The deep learning techniques introduced in this study to estimate groundwater level change under different case scenarios can be convenient for groundwater management as drought warnings and/or water restrictions could be issued in a timely manner. We therefore suggest the use of the modelling framework used as an alternate approach to simulating groundwater level change specifically in areas where subsurface properties are not known.