Selection and Quantification of Best Water Quality Indicators Using UAV-Mounted Hyperspectral Data: A Case Focusing on a Local River Network in Suzhou City, China

Hyperspectral imaging performed by Unmanned Aerial Vehicles (UAVs) has proven its potential in environmental surveillances, especially in the field of water quality monitoring. In this study, three polynomial forms of inversion models for six water quality indicators were specified, with different numbers of spectral reflectance (1/2/3) as independent variables. Each model was designed with seven parameters, and the differential evolution algorithm was used to optimize the parameters by minimization of the mean absolute percentage error (MAPE) between the retrieval results and field observations. Hyperspectral data from a (UAV)-mounted imager and the corresponding river water quality measurements were obtained in a case area in Suzhou City, China. Both MAPE and the coefficient of certainty ( R 2 ) are used to evaluate the model performance. All the models are useable, with an MAPE range of 3–18% and an R 2 range of 0.65–0.94, while the retrieval accuracy is more indicator-dependent and two nitrogen-related indicators have the lowest MAPE of around 5%. Considering the MAPE during model training and verification, the two-band model structure is more robust than the single- or three-band structures. It is certain that such a data-driven approach for large-scale, continuous, and multiple-indicator monitoring with considerable accuracy could facilitate water quality management.