Application of machine learning techniques to improve groundwater level predictions and optimization, West Coast, South Africa

Groundwater systems are complex and subject to climate change, abstraction, and land use stresses, making quantifying their impacts on aquifers difficult. Groundwater models aim to balance abstraction and aquifer sustainability by simulating the responses of an aquifer to hydrological stresses through groundwater levels. However, these models require extensive spatial data on geological and hydrological properties, which can be challenging to obtain. To address this issue, data-driven machine learning models are used to predict and optimize groundwater levels using available data. This paper argues that using machine learning to model groundwater level data improves predicting and optimizing groundwater levels for setting up a managed aquifer recharge scheme. The West Coast Aquifer System in South Africa was used as a case study. The neural network autoregression model was used for the analysis. Multiple variables such as rainfall, temperature, and groundwater usage were input parameters in the mode to facilitate predictions. Outputs from the model showed how machine learning models can enhance the interpretation of observed and modelled results on groundwater levels to support groundwater monitoring and utilization. In areas with high dependence on groundwater and where data on abstraction (use) and monitoring were scarce, results showed that feasible measures were available to improve groundwater security. Although the simulation results were inconclusive, the results provided insights into how the use of machine learning can provide information to inform setting up a managed aquifer recharge scheme.