Chemometrics. Essential tool before, in, and after spectroscopy

Complete text of publication follows. A wide variety of spectroscopic methods is used to solve very complex analytical tasks in research, development, production, and quality assurance. The subjects of study are often characterized by many interesting features. The features to be analyzed, very often have multidimensional characteristics interacting significantly with each other. Owing to this fact and the inhomogeneous distribution of analytes, analysis requires many samples and investigations of several features in parallel. Many of the wide range of spectroscopic methods are useful tools for multifeature analysis, i.e. multielement or multicomponent, respectively. Analysis of mixtures and measurements in the range of traces or even ultra traces are very important fields of research in spectroscopy. On the one hand, these characteristics offer an excellent opportunity to investigate complex systems. However, an enormous flood of complex information, which contains both high variability and high uncertainty to some extent, is generated. Additionally, due to economic and time constraints an analyst's aim is to measure as few features as possible in the minimum of samples. That means e.g. an efficient number of samples by means of optimized measurement procedures. All these issues require chemometric approaches, i.e. mathematical and statistical methods to design and to select optimal experimental/analytical procedures in order to gain maximum relevant information from the data. First, a short overview of methods and goals of chemometrics is presented. Then, useful methods of chemometrics used before spectroscopic measurements will be introduced. Here, the case study 'determining the uncertainty of sampling' will be discussed in detail. It was shown that variance of different steps of the analytical process can be resolved and, thus, can be estimated quantitatively by analysis of variance. Then, distinctive approaches of chemometrics in spectroscopy will be demonstrated and a case study of NIR spectroscopic analysis of starch in rye is discussed. It is demonstrated that the combined approach of NIR analysis and chemometric evaluation using PLS regression is much faster than any common enzymatic starch determination. Another very important field of work in chemometrics deals with the treatment, evaluation, and interpretation of spectroscopic data, i.e. its application after spectroscopy. An overview of the manifold approaches of multivariate data analysis ranging from methods of unsupervised learning over classification methods to factorial methods will be given. Selected methods of data analysis will be explained by case studies from the field of environmental research: evaluation and interpretation of river sediment data. Cluster analysis can be used clearly to detect the multivariate structure of the river course. Factor analysis helps uncover hidden origins of pollution, like dischargers or diffuse effluents. Finally, PLS regression enables the quantitative description of deposition-remobilisation processes between different compartments of a river.