As global industrialization, urbanization, and agricultural activities continue to advance, both natural and engineered water systems are facing severe environmental and operational pressures.

Natural water systems, such as rivers, lakes, and groundwater, are subject to multiple disturbances, including eutrophication, heavy metal accumulation, and nonpoint source pollution, while in engineered water systems, issues such as aging water supply and drainage infrastructure, rising water quality risks, and increasingly complex operational controls have become prominent.

Traditional water environment research and management methods rely on mechanistic and statistical models, as well as manual sampling and analysis. These methods face limitations such as strong parameter dependency, high calibration costs, heavy computational overhead, and insufficient spatial-temporal representation, making it difficult to meet the urgent demands for accurate prediction, real-time perception, and intelligent decision-making in complex water systems.

In this context, machine learning, with its outstanding advantages in non-linear mapping, multi-source heterogeneous data fusion, and high-dimensional feature mining, is becoming a key technical path for transforming the research paradigm of water systems.

To address these challenges, Professor Xu Hao's team at Xi'an Jiaotong University (XJTU) has conducted a systematic review of machine learning applications in natural and engineered water systems, organizing research progress across typical scenarios, including rivers, lakes, groundwater, wastewater treatment plants, and water supply networks.

This study highlights representative achievements in water quality parameter prediction and comprehensive environmental assessment; ecological risk diagnosis and pollution source apportionment, groundwater and complex process simulation; and material and process design, process control optimization, and intelligent operation and maintenance of urban pipe networks.

Based on this, the paper proposes a model selection framework that considers data characteristics, physical constraints, and deployment requirements. It emphasizes that machine learning in water systems is not a simple replacement for traditional mechanistic models, but rather an informed choice among mechanistic models, pure data-driven models, and hybrid models based on the specific problem at hand.

The research further points out that the future development of machine learning in water systems should move from proof-of-concept to trustworthy deployment. Key focus areas include the fusion of physics-informed and explainable machine learning, the combination of digital twins and reinforcement learning, and the application of graph neural networks (GNNs) and federated learning in complex networked water systems.

These advancements will provide theoretical support and methodological references for "smart water" and sustainable water resource management.

The research findings, Machine learning paradigms in natural and engineered water systems: From proof-of-concept to trustworthy deployment, have been published in the prestigious international journal Water Research.