Estimation of surface water quality parameters based on hyper-spectral and 3D-EEM fluorescence technologies in the Ebinur Lake Watershed, China

Abstract Water quality research relies on field sampling, which is often very difficult to obtain, especially in arid areas. This study chose the Ebinur Lake Watershed in arid region as a study area. It analyzed 12 water quality parameters (WQPs) and hyper-spectral derived from 48 field samples. Parallel factor analysis (PARAFAC) method was employed to extract four fluorescent components from the fluorescence excitation-emission matrix (EEM) data. Four fluorescence spectral indices (Fn (355), the fluorescence index (FI), the humification index (HIX)) were used to characterize the organic matter. Estimated WQPs were then coupled with hyper-spectral and three-dimensional fluorescence technologies using the Back Propagation-Artificial Neural Network (BPANN) method developed in this study. The main findings are: (1) Higher correlations exist among the reflectance peaks, spectral indices (DI, NDI, RI) and WQPs, which is helpful to improve the accuracy of water quality estimation. (2) There are also high correlations among fluorescent components (peak) (C1, W2, W3, W4, W5 and W7), fluorescence spectral index and some of WQPs. This indicated that fluorescent components and fluorescence indices can be used to accurately monitor WQPs in surface water. (3) The BPANN model has a great potential for estimating WQPs, because of residual predictive deviation (RPD) of estimation model and verify model more than 1.4. These preliminary results have proved that hyper-spectral and fluorescence technologies is a valuable tool for monitoring surface water quality.