Aquatic system assessment of potentially toxic elements in El Manzala Lake, Egypt: A statistical and machine learning approach

• Developed an Aquatic Water Quality Index (AWQI) for Manzala Lake, Egypt. • Applied four pollution indices: HPI, PI, CI, and MI, for water quality assessment. • Leveraged PCA and CA to refine indices and understand geo-environmental factors. • ANN model disclosed highest prediction accuracy (R² = 1.0) for water quality. • Identified Cd and Cu as key pollutants, impacting lake ecosystem health. This study aimed to assess and predict the surface water quality of Manzala Lake, Egypt, and identify the geo-environmental factors affecting its ecosystem. An Aquatic Water Quality Index (AWQI) was developed alongside four pollution indices (PIs): Heavy Metal Pollution Index (HPI), Pollution Index (PI), Contamination Index (CI), and Metal Index (MI). These indices were refined using multivariate techniques, such as Principal Component Analysis (PCA) and Cluster Analysis (CA). Additionally, six machine learning models, including Multiple Linear Regression (MLR), Decision Tree (DT), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Adaptive Boosting Regression (AdaBoost), and Multilayer Perceptron (MLP), were developed to predict water quality parameters. Water samples, collected over the 2020/2021 and 2021/2022 periods, included sixteen physicochemical parameters for index calculation. The AWQI result of 98.79 indicated severe pollution, rendering the lake’s surface waters unsuitable and detrimental to the aquatic environment. The HPI, MI, and CI values were 88.84, 4.22, and -3.78, respectively, showing varying contamination levels. Cadmium (Cd) and copper (Cu) had the most significant effects, with moderate contributions from zinc (Zn), nickel (Ni), chromium (Cr), lead (Pb), and manganese (Mn), and minor impacts from iron (Fe). The outcomes demonstrated that MLP and DT are the top-performing models for predicting water quality indices. MLP demonstrated exceptional performance for MI (validation RMSE = 0.06, R² = 0.99) and CI (RMSE = 0.13, R² = 0.99). Meanwhile, DT achieved optimal validation results for AWQI (RMSE = 6.29, R² = 0.95) and HPI (RMSE = 9.72, R² = 0.93). Spatial distribution maps revealed that pollution was most severe in the southeastern region of the lake, near the Bahr El-Baqar drain, due to untreated wastewater discharges. Multivariate statistical analysis identified nutrient loading and industrial discharges as the primary drivers of water quality degradation.