Application of hyperspectral analysis of chlorophyll a concentration inversion in Nansi Lake

• Published in: Ecological informatics Vol. 64; p. 101360
• Main Authors: Cui, Yu, Meng, Fei, Fu, Pingjie, Yang, Xinyue, Zhang, Yuxuan, Liu, Pudong
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.09.2021
• Subjects: algorithms; aquaculture; aquatic ecosystems; chlorophyll; Chlorophyll a concentration inversion; Empirical mode decomposition; hyperspectral imagery; Hyperspectral remote sensing; lakes; linear models; Nansi Lake; prediction; reflectance; remote sensing; water quality; XGBoost; Chlorophyll a concentration inversion; Empirical mode decomposition; XGBoost; Hyperspectral remote sensing; Nansi Lake
• ISSN: 1574-9541
• DOI: 10.1016/j.ecoinf.2021.101360

Chlorophyll-a (Chl-a) is an important water quality safety evaluation index, and accurate Chl-a concentration monitoring is important for the development of aquaculture, aquatic ecosystem balance, and drinking water safety. Rapid and accurate Chl-a concentration determination in water using hyperspectral remote sensing is an important subject in water ecological environment monitoring. In this study, the spectral reflectance and Chl-a concentration of Nansi Lake were measured, and the time-frequency method of empirical mode decomposition (EMD) analysis was used for the noise reduction and reconstruction of the first-order differential of the spectrum to extract sensitive spectral features. The eXtreme Gradient Boosting (XGBoost) machine learning algorithm was used to establish a Chl-a concentration estimation model, and the best parameters and model combinations for the inversion of the Chl-a concentration in the water column of Nansi Lake were determined. The results show that the combined three-band algorithm combination parameters obtained from the EMD noise-reduced reconstruction of spectral first-order differential (OFODSR-D) data fit the measured Chl-a concentrations better than the original spectral (OSR) and OFODSR data, with a maximum correlation coefficient of 0.8588. Second, the models based on OFODSR-D achieved more satisfactory prediction results, with XGBoost having the highest estimation accuracy (R2 of 0.9024 and root-mean-square error (RMSE) of 1.1312 μg·L−1 for the inverse model), followed by the partial least squares regression (PLSR) model and the linear model (R2 of 0.8474 and 0.8326, and RMSE of 13.3031 and 7.6987 μg·L−1, respectively). This study innovatively introduces the EMD method to the spectral processing of water bodies, obtains optimal parameters for the inversion of the Chl-a concentration, and achieves better results. This study provides a new approach to obtaining optimal inversion parameters for Chl-a monitoring in inland lake water bodies using remote-sensing methods. •The Vis- NIR spectra were used to estimate Chl-a concentration in Nansi Lake water body.•The frequency domain analysis method of EMD could be used for water spectrum processing.•The combination of optimized parameter and model were founded for inversion of Chl-a concentration.