Artificial Neural Network Technique for Predicting Groundwater in Bida Basin, Mokwa, Niger State, Nigeria

Abstract

In this paper, groundwater level during the dry season in Bida Basin, Mokwa is predicted. An Artificial Neural Network (ANN) was applied to investigate the practicability of mass balance equation for the network training and testing. The Feed Forward Levenberg Marquardt (FFLM), Recurrent Neural Network and cascade with Resilient Back propagation for different algorithms was used to calculate groundwater levels from October to April which is dry season in Mokwa, Bida Basin. The performance of the models was evaluated using Mean Square Error (MSE) and Correlation Coefficient. Two lithological group: unconfined and Semi-confine were considered and Climatic data from January 2013 to December 2018 was used for the network training and testing. The results show that the Feed Forward Levenberg Marquardt (FFLM) is the best overall performance for groundwater prediction in Mokwa with Mean Square Error (MSE) of 3.94 and corresponding correlation coefficient of 0.85, these means that the depth to groundwater levels in Mokwa increases from December and reached its highest level in April and reached its lowest level in September. It is observed that Actual Groundwater level in Mokwa is 27.19018 million cubic m for the total of 15 hectares and the Predicted Groundwater level is 26.7889 million cubic for the total of 15 hectares while the difference between Actual Groundwater level and Predicted Groundwater level is 0.40128. Artificial Neural Network (ANN) techniques were well suited for groundwater prediction level.