Eutrophication Inversion of Reservoir Waters Based on Deep Learning and UAV Hyperspectral Remote Sensing

Abstract

Eutrophication of reservoir water poses significant risks to human life and production, necessitating large-scale, regular monitoring and early warning. This work comprehensively evaluated the eutrophication status of reservoir water quality using four water quality parameters: chlorophyll a (Chl-a), Total Phosphorus (TP), Total Nitrogen (TN) and Permanganate Index (COD_Mn). A water quality parameter inversion scheme based on deep learning and UAV hyperspectral remote sensing images is proposed. Based on the optimized feature combinations and sample sets, a water quality parameter concentration regression prediction model based on deep convolutional residual neural network (WQR-DCRNN) is designed and constructed for the inversion of the water quality parameters of Chl-a, TP, TN, and COD_Mn. Then, the parameters of above WQR-DCRNN models are optimized by reinforcement learning. Taking Shanzai Reservoir in Fuzhou City as an example, the average inversion accuracy of the four water quality parameters reached 0.91, 0.80, 0.81, and 0.78, respectively, and the mean relative error (MRE) can meet the requirement of the industrial standards for spectral water quality detection. The concentration spatial distribution patterns of the four water quality parameters provided by this work can provide useful decision-making references for the assessment, prevention and control of eutrophication in large-scale reservoir water bodies.