A neural-network model to retrieve CDOM absorption from in situ measured hyperspectral data in an optically complex lake: Lake Taihu case study

Abstract

Coloured dissolved organic matter (CDOM) is an important water component that affects water colour and ecological environment under water. The remote estimation of CDOM is always a challenge in the field of water-colour remote sensing owing to its weak signal. To further study the CDOM-retrieval approach, field experiments, including water-quality analysis and spectral measurements, were carried out in Lake Taihu waters from 8 to 21 November 2007. On the foundation of analysing water-inherent optical properties, sensitive spectral factors were selected, and then neural-network models were established for retrieving CDOM. The results show that the model with 10 nodes in the hidden layer performs best, yielding a correlation coefficient (R) of 0.887 and a root-mean-square error of 0.156 m⁻¹. Meanwhile, the predictive errors of the model developed here and the previously proposed algorithms were compared with each other. The mean value of the relative error of the former is 12.8% (standard deviation of 29.9%), and is much lower than its counterpart of other models, which indicates that the developed model has a higher accuracy for CDOM retrieval in Lake Taihu waters. Meanwhile, other datasets collected at different times were also imported into the model for applicability analysis; the derived errors suggest a relatively good performance of the model. This research firstly explores the CDOM retrieval in optically complex lake waters, and the corresponding findings support a technical framework for accurately extracting CDOM information in Lake Taihu waters, based on an adequate understanding of water optical properties.

Acknowledgements

This research was supported by a Science Research Starting Fund (No. 20100410) of Nanjing University of Information Science and Technology, a National Natural Science Foundation (No. 40971215), and a National Doctoral Foundation (No. 20093207110011) of China. Kun Shi, Yu Yang, Rui Xia, Xin Jin, Yanfei Wang, Bing Yin, Hong Zhang, Yifan Xu, Zhonghua Liu and Xin Xu are acknowledged for their participation in the field experiment. We would like to express our deep thanks to Dr. Li Lin from Indiana University for checking the English language, and two anonymous reviewers for their useful comments and suggestions.