Predicting groundwater drawdown in Zakho region, Northern Iraq, using machine learning models optimized by the whale optimization algorithm

Abstract

Predicting groundwater drawdown is crucial to the Duhok Governorate’s sustainable management of its water resources. To ensure long-term water availability as extraction from population growth and development intensifies, predicting drawdown helps to prevent overuse, provide a continuous supply of water, and enable effective planning for urbanization, agriculture, and industrial needs. In this work, a novel approach based on Multi-layer perceptron neural network (MLP), support vector regression (SVR), k-nearest neighbor algorithm (KNN), and extreme learning Machine (ELM) optimized by whale optimization algorithm (WOA) were proposed for estimating the total drawdown at Zakho region, Duhok Governorate, Northern Iraq for the first time. The input variables of the models include the rate of water extraction from the well (Q), well depth (D), and various meteorological parameters such as rainfall (R), evapotranspiration (E), Maximum Temperature (Tmax), and Minimum Temperature (Tmin). It is found that ELM showed the highest performance in modeling groundwater drawdown (R² = 0.911, RMSE = 5.674 m, and MAE = 4.937 m). Moreover, the novelty of the research work is to enhance the accuracy of the individual models using two ensemble techniques including simple averaging ensemble (SAE) and weighted average ensemble (WAE). Based on the findings, the WAE technique increased the performance of individual models by up to 20%, proving the reliability of the WAE technique for groundwater drawdown prediction.