Chemometric methods and near-infrared spectroscopy applied to bioenergy production

The present work examines bioenergy production from different viewpoints. The three main objectives are: (1) to reveal the relation of technology, sustainability and economy in bioenergy processes; (2) to investigate spectroscopic methods as a tool for analytical monitoring of bioenergy processes; and (3) to develop new chemometric methods for advanced analysis of spectroscopic data. At the first stage, this thesis investigates the technological, ecological, and economic features of renewable-resource-based and de-centralized bioenergy production systems. In different scenarios, small-scale bioethanol production is combined with other technologies that provide renewable energy from residuals of the bioethanol process. The general aim is to substitute fossil energy conventionally used within the bioethanol process. The investigated technologies are biogas production and straw incineration. Agricultural aspects are introduced by sustainable crop rotation concepts that reconcile food, feed, and biofuel production. The sustainability of small-scale bioethanol production in the different scenarios is quantified by an ecological footprint method, the sustainable process index, SPI, and compared to conventional fuels. The main findings are: (i) small-scaled bioethanol production can be operated with 100 % renewable energy supply, (ii) the SPI of bioethanol can be reduced up to 92 % compared to conventional fuels, (iii) a complex trade-off between ecology-of-scale and economy-of-scale is necessary. At the second stage, this thesis approaches bioenergy production processes from an analytical perspective, and presents near-infrared spectroscopy (NIR) as promising method for fast process monitoring of bioethanol production and biomass characterization. In addition, new analytical methods are presented for a fast determination of the heating value of solid biomass fuel, based on IR and NIR spectroscopy. The main findings are that NIR spectroscopy and appropriate chemometric data analysis (i) successfully determine the concentrations of moisture, protein, and starch in the feedstock material as well as glucose, ethanol, glycerol, lactic acid, acetic acid in the processed bioethanol broths; (ii) and allow quantifying a complex biofuel's property such as the heating value. At the third stage, this thesis focuses on new chemometric methods that improve mathematical analysis of multivariate data such as NIR spectra. The newly developed method 'repeated double cross validation' (rdCV) separates optimization of regression models from tests of model performance; furthermore, rdCV estimates the variability of the model performance based on a large number of prediction errors from test samples. The rdCV procedure has been applied to both the classical PLS regression and the robust 'partial robust M' regression method, which can handle erroneous data. The peculiar and relatively unknown 'random projection' method is tested for its potential of dimensionality reduction of data from chemometrics and chemoinformatics. The main findings are: (i) rdCV fosters a realistic assessment of model performance, (ii) robust regression has outstanding performance for data containing outliers and thus is strongly recommendable, and (iii) random projection is a useful niche application for high-dimensional data combined with possible restrictions in data storage and computing time. The three chemometric methods described are available as functions for the free software R. (author)