The Application of Machine Learning for Groundwater Level Prediction in the Steenkoppies Compartment of the Gauteng and North West Dolomite Aquifer, South Africa

Groundwater in the Steenkoppies compartment of the Gauteng and North West dolomite aquifer is extensively used for agricultural practices that can potentially lead to groundwater storage depletion, threatening groundwater sustainability in the compartment. Groundwater levels represent the response of an aquifer to changes in storage, recharge, discharge, and hydrological stresses. Therefore, groundwater levels are useful for identifying limits and unacceptable impacts on an aquifer and using this information to implement sustainable groundwater management decisions. The use of machine learning techniques for groundwater modelling is relatively novel in South Africa. Conventionally, numerical techniques are used for groundwater modelling. Unlike traditional numerical models, machine learning models are data-driven and learn the behaviour of the aquifer system from measured values without needing an understanding of the internal structure and physical processes of an aquifer. In this study, Neural Network Autoregression (NNAR) was applied to obtain groundwater level predictions in the Steenkoppies compartment of the Gauteng and North West Dolomite Aquifer in South Africa. Multiple variables (rainfall, temperature, groundwater usage and spring discharge) were chosen as input parameters to facilitate groundwater level predictions. The importance of each of these inputs to aid the prediction of groundwater levels was assessed using the mutual information index (MI). The NNAR model was further used to predict groundwater levels under scenarios of change (increase or decrease in recharge and abstraction). The results showed that the NNAR could predict groundwater levels in 18 boreholes across the Steenkoppies aquifer and make predictions for scenarios of change. Overall, the NNAR performed well in predicting and simulating groundwater levels in the Steenkoppies aquifer. The transferability of the NNAR to model groundwater levels in different aquifer systems or groundwater levels at different temporal resolutions requires further investigation to confirm the robustness of the NNAR to predict groundwater levels.