Modeling groundwater quality and prediction in a highly urbanised coastal city of Lagos using an artificial neural network

Abstract

The quality of groundwater is a serious issue worldwide. The groundwater quality status was assessed by comparing it with World Health Organization (WHO) standards, and groundwater quality parameters were predicted using an Artificial Neural Network (ANN) model with the Levenberg-Marquardt algorithm. A total of 15 samples were collected from boreholes and wells and analyzed for Electrical Conductivity (EC), pH, temperature, salinity, and chloride, and predictions were made for Total Dissolved Solids (TDS), sodium (%Na), iron (Fe), and bicarbonate. The data were normalized and then divided into training, validation, and testing sets (70%, 15%, and 15%, respectively). The results indicated significant variations in the spatial distribution; Isheri had significantly higher EC (5860 µS/cm) and TDS (3180 mg/L), which are beyond WHO limits, likely due to industrial pollution. The salinity level on Tincan Island was high (2.24 ppt), and the water was undrinkable, whereas the salinity levels on Akodo and Akeusola were low (0.1 to 0.08 ppt), and the water was drinkable. The ANN had good predictive performance with Mean Square Error of 0.6908 (training), 0.5063 (test) and 1.11 (validation). The regression value of the 10-neuron hidden layer was 1.0 (training), 0.92601 (validation), and 0.95023 (testing). Absolute Error was minimal (TDS: 0.0 to 0.06; Fe: 0.0 to 0.001), and Mean Absolute Error (TDS: 0.84; Na: 0.17; Fe: 0.0019; bicarbonate: 0.5973), confirming the model's reliability. The study shows that ANN is a powerful tool for predicting groundwater quality and can support sustainable groundwater management in coastal urban areas. The disadvantages are a small sample size (15) and the use of fixed parameters that do not account for seasonal changes. Larger data sets and dynamic environmental factors should be included in future research.