Advances in Water Quality Forecasting Using Deep Learning: A Review of Fundamental, Models, Evaluation and Challenges

Abstract

Forecasting of water quality is crucial for safeguarding public health, optimizing resource management, and promoting environmental sustainability. However, predicting water quality dynamics is challenging due to nonlinear relationships among environmental variables, temporal fluctuations, and data scarcity. Recent advances in deep learning have significantly enhanced the predictive accuracy of water quality indices (WQIs) and associated parameters. This review critically examines deep learning approaches for water quality forecasting, focusing on surface water. Following a systematic literature search covering publications from 2015 to June 2026 and a detailed eligibility audit of all screened references, 37 primary studies were included in the systematic synthesis, supplemented by additional review, methodological, and contextual references cited for background. The architecture reviewed includes convolutional and recurrent neural networks, hybrid models, attention mechanisms, transformers, and graph neural networks for spatial and temporal modeling. Data sources examined include in situ sensors, laboratory analyses, remote sensing, and hydro-meteorological variables. Prevalent challenges such as missing data, non-stationarity, and spatial limitations are addressed. Mitigation strategies, including transfer learning, uncertainty quantification, and physics-informed approaches are synthesized. The key finding is that hybrid and attention-enhanced deep learning architectures are frequently reported to outperform single architecture baselines in the reviewed literature and offer the greatest promise for robust, interpretable, and scalable systems for predicting water quality.