New approaches for surface water quality estimation in Lake Erken, Sweden, using remotely sensed hyperspectral data

This work demonstrates the efficiency of using linear statistical modelling for estimation of concentrations of various substances in lake water using remotely sensed multiand hyperspectral images together with extensive field measurements collected over Lake Erken in Sweden. A linear relationship was assumed between image data and the corresponding field measurements, and the transformation coefficients were estimated using the least squares method. The resulting coefficients were used to transform new image data into the corresponding substance concentrations. Estimation errors were computed and concentration maps were generated for chlorophyll-a and phaeophytine-a, suspended particulate organic matter SPOM, suspended particulate inorganic matter SPIM, as well as total suspended particulate matter SPM (SPOM+SPIM). Good correlation was obtained between estimated and measured values. Backward elimination was performed to find the most useful spectral bands for the case study of this work. Descriptive spectral signatures, describing the impact of underlying processes on the spectral characteristics of water, were generated, analysed and also used to predict the corresponding water quality parameters in image data, with the same estimation accuracy as the linear statistical model. Feature vector based analysis FVBA was also employed to generate transformation coefficients that could be used to estimate water quality parameters from image data, also, with the same accuracy as the previous methods. Finally, the impact of performing atmospheric correction was investigated, in addition to applying linear statistical modelling for the purpose of combined atmospheric correction and ground reflectance estimation. Key-Words: Remote sensing, Statistical modelling, Water quality, Descriptive spectral signatures