1. School of Soil and Water Conservation, Beijing Forestry University, Beijing 100038, China
2. Key Laboratory of Soil and Water Conservation and Desertification Prevention, Beijing Forestry University, Beijing 100083, China
Abstract：With the rapid development of industrialization and social economy, water pollution and deterioration of water sources are increasingly aggravated, and effective water quality monitoring is an important prerequisite for water source protection. Miyun Reservoir is an important surface water source in Beijing, which plays an important role in protecting water safety in the capital. In order to monitor the water quality parameters and pollution degree of Miyun Reservoir more accurately, this study used four phases of UAV hyperspectral remote sensing data to construct a water quality parameter retrieval model based on a deep neural network algorithm. Total nitrogen (TN) and total phosphorus (TP) water quality parameters in Miyun Reservoir were retrieved. Firstly, the hyperspectral image dimensionality reduction processing based on the recursive feature elimination method was used, and the spectral data and groundwater quality monitoring data were superimposed. The network structure parameters, such as the number of hidden layers and the number of ganglion points, were determined by minimizing the error in the training process. Then, the migration method gradually expanded the network from knowledge source domain to network, and the water quality parameters of TN and TP concentration in Miyun reservoir were trained and verified. Finally, the water quality parameters of Chaohe Dam and Baihe Dam in Miyun Reservoir were retrieved to reveal the spatio-temporal evolution of the main water quality parameters. The results show that ① the R2 of the TN and TP concentration retrieval models constructed in this study are 0.835 5 and 0.770 3, and the MSE is 0.015 3 and 0.000 8. The Ensemble Deep Belief Network (EDBN) model based on random subspace has a better retrieval effect on water quality parameters. ②TN concentration in Miyun Reservoir fluctuates with seasons, with a low concentration in summer and a relatively high concentration in autumn. The change in TP concentration is relatively stable, indicating that the control effect of phosphorus pollution in the surrounding area of Miyun Reservoir is good.③The water quality of the Baihe Dam was better than that of the Chaohe Dam. The seasons obviously affected the changes of the former, while the latter was significantly affected by human activities. The TN concentration of Miyun reservoir was in Class III, and the TP was generally in Class II. The water quality can meet the standards of drinking water sources, but it is still necessary to strengthen the supervision of nitrogen and phosphorus pollution. These results will provide an important scientific basis for efficiently monitoring water quality and water resources protection in the Miyun reservoir.
Key words：UAV hyperspectral; Deep neural network algorithm; Water quality retrieval; Miyun Reservoir
乔 智，姜群鸥，律可心，高 峰. 无人机高光谱遥感和集成深度置信神经网络算法用于密云水库水质参数反演[J]. 光谱学与光谱分析, 2024, 44(07): 2066-2074.
QIAO Zhi, JIANG Qun-ou, LÜ Ke-xin, GAO Feng. Retrieval Model for Water Quality Parameters of Miyun Reservoir Based on UAV Hyperspectral Remote Sensing Data and Deep Neural Network
Algorithm. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2024, 44(07): 2066-2074.
