Retrieving Eutrophic Water in Highly Urbanized Area Coupling UAV Multispectral Data and Machine Learning Algorithms

Abstract

With the rapid development of urbanization and a population surge, the drawback of water pollution, especially eutrophication, poses a severe threat to ecosystem as well as human well-being. Timely monitoring the variations of water quality is a precedent to preventing the occurrence of eutrophication. Traditional monitoring methods (station monitoring or satellite remote sensing), however, fail to real-time obtain water quality in an accurate and economical way. In this study, an unmanned aerial vehicle (UAV) with a multispectral camera is used to acquire the refined remote sensing data of water bodies. Meanwhile, in situ measurement and sampling in-lab testing are carried out to obtain the observed values of four water quality parameters; subsequently, the comprehensive trophic level index (TLI) is calculated. Then three machine learning algorithms (i.e., Extreme Gradient Boosting (XGB), Random Forest (RF) and Artificial Neural Network (ANN)) are applied to construct the inversion model for water quality estimation. The measured values of water quality showed that the trophic status of the study area was mesotrophic or light eutrophic, which was consistent with the government’s water-control ambition. Among the four water quality parameters, TN had the highest correlation (r = 0.81, p = 0.001) with TLI, indicating that the variation in TLI was inextricably linked to TN. The performances of the three models were satisfactory, among which XGB was considered the optimal model with the best accuracy validation metrics (R2 = 0.83, RMSE = 0.52). The spatial distribution map of water quality drawn by the XGB model was in good agreement with the actual situation, manifesting the spatial applicability of the XGB model inversion. The research helps guide effective monitoring and the development of timely warning for eutrophication.

Keywords

• eutrophic water
• UAV remote sensing
• machine learning
• water quality inversion
• extreme gradient boosting