作者：Shi, J., Li, Z., Liu, G., Fang, C., Wen, Z., Shang, Y., Tao, H., Li, S., Wang, Q., Wang, Z. & Song, K.

A combination of Machine Learning (ML) techniques with satellite remote sensing has brought target objects and water-quality research into a new era. These emerging techniques and unprecedented opportunities have improved the accuracy of satellite data products, facilitated knowledge discovery, expanded the spatial domain of target objects and water-quality research to the globe, and extended the social, economic, and scientific impacts of water-quality research. This review provides the first systematic integration of machine learning (ML) across the full-chain workflow for inland and nearshore watercolor remote sensing, we examine recent progress in the application of ML across the full remote sensing workflow for inland and nearshore water color studies-from image preprocessing (including atmospheric correction, water body extraction, spectral simulation, spatiotemporal fusion and water extent extraction) to object detection and water quality parameter estimation. We also present a curated selection of valuable open-source resources and highlight key challenges-including data scarcity and label bias, limited model interpretability, insufficient uncertainty quantification, and the difficulties of multimodal data integration, long-term, large-scale, and multi-scale analysis, as well as the need for effective spatial relationship modeling-and discuss emerging trends such as multitemporal data integration, and the incorporation of physical constraints. By providing a comprehensive overview of ML applications in this field, this review serves as a valuable reference for researchers and practitioners seeking to enhance water quality monitoring and assessment. The insights presented here are expected to foster the development of more accurate, scalable, and interpretable ML models, ultimately contributing to improved environmental management and decision-making.

机器学习（ML）技术与卫星遥感的结合，将目标地物与水质研究带入了新纪元。这些新兴技术与前所未有的机遇提升了卫星数据产品的精度，促进了知识发现，将目标地物与水质研究的空间域拓展至全球，并扩大了水质研究的社会、经济与科学影响。本综述首次系统整合了机器学习在内陆与近岸水色遥感全链条工作流程中的应用：从图像预处理（包括大气校正、水体提取、光谱模拟、时空融合与水体范围提取）到目标检测与水质参数反演，全面梳理了ML在水色遥感各环节的最新进展。我们还精选整理了有价值的开源资源，并指出关键挑战——包括数据稀缺与标签偏差、模型可解释性有限、不确定性量化不足，以及多模态数据整合、长时序/大尺度/多尺度分析困难与有效空间关系建模的需求——同时讨论了多时相数据融合与物理约束嵌入等新兴趋势。通过提供该领域ML应用的全面综述，本文可为致力于提升水质监测与评估能力的研究者与从业者提供有价值的参考。文中提出的见解有望推动更准确、可扩展、可解释的ML模型的发展，最终助力环境管理与决策的改进。

（来源：Journal of Hydrology 2026  DOI: 10.1016/j.jhydrol.2026.135950）