Assessment of CSO loads – based on UV/VIS-spectroscopy by means of different regression methods

Abstract

The use of UV/VIS-spectroscopy for water quality measurements is based on the solution of the correlation between the surrogate parameter absorbance and the resulting equivalence parameters. The coherence of absorbance and equivalence parameters (CODtot, CODsol, TSS) is solved in this paper with different regression methods. The correlation of absorbance and concentrations are analysed based on linear regression methods, model tree regressions, multivariate regression methods and support vector machines using sequential minimal optimisation algorithm. For this purpose the regression methods are calibrated on three 24hours measurement campaigns of a combined sewer measurement station situated in the combined sewer overflow chamber in Graz (Austria). The online measurement station has been conveying data for more than 2½ years up to now. Finally, the load calculation based on the different regression methods and its comparison demonstrate that an apparently complex model does not inevitably lead to accurate concentration values due to possible model overfitting. Hence, the paper points out the possibilities and the drawbacks of spectroscopy measuring in sewers and the arising concentration values.