Evaluating Effective AI Models for Surface and Groundwater Quality Prediction: A Review

Artificial intelligence (AI) techniques have become important tools for predicting and assessing water quality because they can capture nonlinear relationships among complex environmental variables. This review systematically analyzes AI-based approaches for surface water and groundwater quality prediction and identifies factors that influence model performance. Following PRISMA guidelines, we conducted a systematic review using five scientific databases, screening ۲۲۴ records and including ۳۵ peer-reviewed studies. We analyzed the reviewed studies by AI model type, water resource category, input parameters, dataset characteristics, hydrological conditions, and prediction performance. The results indicate that hybrid and ensemble models generally outperform standalone algorithms by improving feature extraction, reducing uncertainty, and better capturing nonlinear relationships. GPR showed high accuracy for WQI prediction with uncertainty below ۲%, RF-SVM achieved strong performance for BOD prediction (R² = ۰.۹۰۸), and LSTM-based and metaheuristic models improved DO prediction. Despite their high accuracy, AI models still face challenges related to interpretability, data availability, and regional transferability. This review compares AI model suitability under different hydrological conditions and proposes future directions in explainable AI, uncertainty quantification, and adaptive real-time water quality monitoring.

کلیدواژه ها:

Artificial Intelligence ، Water quality ، Surface water ، Groundwater ، Water management ، Hydro-informatics ، sustainability