Retrieval model for total nitrogen concentration based on UAV hyper spectral remote sensing data and machine learning algorithms – A case study in the Miyun Reservoir, China

• Main Authors: Huijie, X. (Author), Lidan, X. (Author), Meilin, W. (Author), Qun'ou, J. (Author), Siyang, S. (Author)
• Format: Article
• Language: English
• Published: Elsevier B.V. 2021
• Subjects: Adaptive boosting; Bayesian networks; Beijing [China]; China; Concentration change; Concentration retrieval; Decision tree regression; Decision trees; Dimensionality reduction; Landsat; Learning systems; machine learning; Machine learning algorithm; Miyun Reservoir; Miyun reservoir , China; Neural networks; nitrogen; Nitrogen; reaction kinetics; real time; Remote sensing; remotely operated vehicle; Reservoir management; Reservoirs (water); Retrieval model; River pollution; Rural areas; Scientific references; Spatial distribution; Spatial heterogeneity; Support vector machines; Support vector regression; Total nitrogen concentration; UAV hyper spectral remote sensing data; Unmanned aerial vehicles (UAV); water pollution; Water pollution levels; water quality; Water quality; Water quality monitoring; Water quality parameter
• Online Access: View Fulltext in Publisher
• ISBN: 1470160X (ISSN)
• DOI: 10.1016/j.ecolind.2021.107356

Monitoring the water pollution level in real time is the most critical issue for protecting the water quality of water reservoirs. Due to the restrictions on flight areas of Unmanned Arial Vehicles (UAV), four sensitive regions with the area of 1–2 km2 were first selected in this study based on the spatial distribution of total nitrogen (TN) concentration changes estimated by the Landsat remote sensing data. And then twelve machine learning algorithms including Support Vector Machine (SVM), Artificial Neural Networks (ANN), Bayesian Ridge Regression (BRR), Lasso Regression (Lasso), Elastic Net (EN), Linear Regression (LR), Decision Tree Regression (DTR), K Neighbors Regression (KNR), Random Forest Regression (RFR), Extra Trees Regression (ETR), AdaBoost Regression (ABR) and Gradient Boosting Regression (GBR) were compared to construct a more accurate retrieval model by using the UAV hyper spectral remote sensing and ground monitoring data. And then the TN concentration was estimated after the process of dimensionality reduction and compressed sensing denoing. Finally, spatial heterogeneity of the TN concentration was analyzed in four sensitive areas of the Miyun reservoir. The results demonstrated that among the tested algorithms the Extra Trees Regression was best suitable for the construction of a TN concentration retrieval model on the basis of UAV hyper spectral data, and its absolute squared error was 0.000065. The spatial distribution of the TN showed that the concentration was highest within the water area of the Bulaotun village and the Houbajiazhuang village, while it was relatively low for the Chao river dam and Bai river dam. Additionally, no significant differences regarding the concentrations were shown in the single UAV flight area except the Houbajia village, which indicated that the water quality in Miyun reservoir was relatively stable and changed in a small interval. These conclusions can provide scientific references for water quality monitoring and management in the water reservoir. © 2021 The Author(s)