Neural Network Modeling for Retrieval of Water Quality of Lake Taihu from Field Spectral Measurement

How to effectively monitor water quality of Coastal and inland waters by optical remote sensing has always been a difficulty. This study develops a neural network model to improve the accuracy in monitoring water quality of Lake Taihu, China, a large shallow subtropical lake. A three-layer back-propagation neural network is built up to estimate concentrations of chlorophyll-a, total suspended matter and dissolved organic carbon at the same time from in-situ measured water surface spectra. The neural network is trained by the simultaneously in-situ measured water surface spectra and water quality parameters (including concentrations of chlorophyll-a, total suspended matter and dissolved organic carbon) in 39 sampling stations of Lake Taihu in the campaign in winter of 2007. Then, the neural network model is tested by the data measured in both the left 9 sampling stations in the campaign in winter of 2007 and the 9 sampling stations in the campaign in winter of 2006. The estimated errors of the three kinds of water quality parameters are less than 30% in both campaigns in winters of 2006 and 2007. The results show that this neural network model has high seasonal applicability, and is very useful in monitoring of water quality of Lake Taihu from water surface spectra measured in future winters.