Approximation of Aquifer Replenishment from Rainfall, Streamflow and Base Flow Index using Data-Driven Models

An optimized informed decision making is of great advantage in the water resources management process. Data-driven models have become integral to the development of smart systems and software, that have become part of our life; because of their ability to provide reliable approximate of complex non-linear systems. This study assessed the performance of machine learning algorithms for the approximation of groundwater replenishment into the basement complex aquifer of the Upper-Central Ogun Basin in Southwest Nigeria. The study applied an Artificial Neural Network (ANN), XgBoost and Decision Trees algorithms for predicting groundwater recharge based on input data such as total annual rainfall, streamflow, and base flow index. While the performance of the models was evaluated using Mean Absolute Error (MEA), Root Mean Square Error, Correlation Coefficient (R) and Coefficient of Determination (R2). Analysis of the performance of the three models revealed that based on their ability to predict aquifer replenishment, ANN with a single layer, consisting of eight (8) nodes had an accuracy of 96%, in contrast to XgBoost, which had 20% accuracy and Decision Trees, which had 44% accuracy. This implies that the developed ANN model can be further deployed in the cloud for the development of mobile and web-based applications, to enhance fast and reliable informed decision making in the management of water resources in the Ogun River Basin and Nigeria. The study concludes that a single hidden-layered ANN model with eight nodes is adequate for approximating aquifer replenishment in the study area.