Developing Water Quality Indices Utilizing Artificial Neural Networks: A Case Study of the “Gharbia” Main Drain in Egypt

Egypt faces significant challenges in managing its limited water resources, with agricultural drainage water merging as a crucial unconventional water resource for irrigation. This study focuses on the Gharbia Main Drain, a vital water source providing 1.9 billion cubic meters of water annually, but heavily polluted by agricultural runoff, domestic wastewater, and industrial discharges. To address this, Water Quality Indices (WQIs) were developed using Artificial Neural Networks (ANNs) to assess pollution levels. Nineteen water quality parameters were analyzed, and the Canadian Water Quality Index (CWQI) was calculated. The dataset was preprocessed for training ANN models. Results were analyzed annually and seasonally, supported by sensitivity analysis and feature importance to identify key parameters influencing WQIs. Five WQI models were developed to evaluate water quality based on pollution types: biological, industrial, and agricultural pollution types. Biological pollution was identified as the main contributor to contamination in the drain. Water quality progressively improved from the inlet to the outfall, with downstream branch drains showing better water quality than upstream branches, suggesting potential reuse of downstream water before it flows into the main drain. The results of the feature importance and sensitivity analysis identified seven key parameters: Fecal Coliform (FC), Biological Oxygen Demand (BOD), Chemical Oxygen Demand (COD), Dissolved Oxygen (DO), Ammonium (NH₄), Total Dissolved Solids (TDS), and Electrical Conductivity (EC) as the most influential when evaluating WQIs. ANN models demonstrated superior reliability and predictive accuracy compared to traditional methods, effectively capturing nonlinear data relationships and offering robust tools for sustainable water management.