Groundwater Level Forecasting Using Multiple Linear Regression and Artificial Neural Network Approaches

Abstract

Accurate and reliable groundwater level prediction is a critical component in water resources management. This paper developed two methods to predict forty-six months of groundwater level fluctuation. The approaches of Multiple Linear Regression (MLR) and Artificial Neural Network (ANN) were compared for predicting groundwater levels. MLR and ANN approaches were performed at two monitoring wells, Ubung and Ngurah Rai, in the Denpasar region of Bali, Indonesia, considering all significant inputs of hydrometeorological time series data: barometric pressure, evaporation, temperature, wind, bright sunshine, rainfall, and groundwater level. The model's performance was assessed statistically and graphically. The ANN-predicted groundwater levels agreed better with the observed groundwater levels than the MLR-predicted groundwater levels at all sites. The results show the ANN performs better than MLR in terms of statistical errors, notably mean square error (MSE) value of 0.6325; root mean square error (RMSE) value of 0.7953; mean absolute error (MAE) value of 0.6122 based on the MLR in the Ubung monitoring well, while ANN models got an MSE value of 0.143; RMSE value of 0.379, and MAE value of 0.311. For the Ngurah Rai monitoring well, the MSE value is of 1.3406, RMSE value of 1.1579, and MAE value of 0.9152 for MLR, while ANN models obtained MSE value of 0.0483, RMSE value of 0.2198, and MAE value of 0.1266.