High-precision inversion of urban river water quality via integration of riparian spatial structures and river spectral signatures.

With the ongoing process of urbanization, it poses challenges to the monitoring of water quality in urban rivers. The mainstream methods for remote sensing water quality monitoring rely on the optical characteristics of water to achieve water quality inversion, while overlooking the correlation between water quality and riparian zones. The spatial arrangement and scale fluctuation of the riparian zones exert a substantial influence on water quality as it serves as an intermediary region connecting riverine and terrestrial ecosystems. Therefore, this study firstly employed unmanned aerial vehicle (UAV)-borne multispectral remote sensing technology to capture the subtle variations in urban river water quality and obtain detailed spatial information of the riparian zone. The Liang-Kleeman information flow was subsequently employed to quantitatively assess the causal responses of the spatial composition of riparian zone to water quality parameters across various spatial scales. Finally, we developed a hierarchical ensemble learning model for water quality assessment by integrating the spatial characteristics of the riparian zone with the spectral properties of the water body. The result demonstrates that this model accurately delineated water quality grades for three key parameters: ammonia nitrogen (NHN), chemical oxygen demand (COD), and total phosphorus (TP), achieving accuracies of 94.87 %, 92.31 %, and 89.74 %, respectively. Our study presents a water quality inversion method for urban rivers, which holds significant guidance for the monitoring and management of urban rivers and contributes to further promoting the sustainable development of cities.