Deep learning-based remote sensing retrieval of inland water quality: A review
Zhiguo Pang, Zhuoyue Zhou, June Fu, Wei Jiang, Xiangdong Qin, Minghan Sun
Abstract
Study region This review focuses on inland waters worldwide. Study focus Inland water quality monitoring is essential for ecological protection and sustainable water resource management. Advances in remote sensing and deep learning have enabled the retrieval of water quality parameters by combining large scale observations with the modeling of complex spectral and spatial patterns. This review examines key advances in feature construction, model architectures, and optimization for deep learning-based inland water quality retrieval. Input features have evolved from simple spectral vectors to spatial patches and fused data from multiple sources, enabling richer representations of water environments. Model architectures range from fundamental structures to hybrid networks integrating sequence modeling, attention mechanisms or autoencoder-based feature compression. Various optimization strategies have been proposed to improve model generalization and interpretability, including physically informed losses, transfer learning, and uncertainty quantification. New hydrological insights for the region Future research should focus on building systematic frameworks for feature selection and integration, with particular attention to adaptive patch sizing, multi-source data assimilation, and expanding sequence modeling beyond time series to include spectral band sequences at the pixel level. Further efforts are needed to enhance hybrid modeling through multi-scale attention and physical constraints, incorporate uncertainty feedback into dynamic optimization, and establish benchmarks for model stability and regional transferability.