Setup and optimization of a PLS regression model for predicting element contents in river sediments

A partial least-squares (PLS) regression model was used to predict element contents in sediments of the river Elbe from measured contents in the particulate suspended matter (SPM). This paper shows how to preprocess the data and to find the most suitable prediction variables for this problem by means of a simulated annealing-based optimization algorithm. Special emphasis is also laid on data postprocessing and verifying the quality of the predictions. Thus, a regression model could be proposed to predict the contents of at least 12 selected elements in the sediment with the aid of 30 measured element contents in the suspended matter with a good predictive quality.