Speaker
Description
Spectral comparison is central to many analytical decisions, including material identification, library searching, batch comparability, forensic screening, process monitoring and waste sorting. However, the apparently simple question of whether two spectra are similar can be answered in mathematically different ways. This presentation introduces the fundamentals of chemometric spectral comparison in infrared spectroscopy, with emphasis on the rationale behind commonly used distance- and similarity-based metrics. Classical Manhattan and Euclidean distances are presented as point-to-point measures of spectral deviation, while weighted spectral distance approaches are discussed as strategies to emphasise chemically informative regions and, when reference-class variability is considered, to account for spectral uncertainty. These methods are contrasted with similarity indices such as cosine similarity, Pearson and Kendall correlations, area of overlap and normalised local change, each reflecting a different definition of resemblance based on vector orientation, linear shape agreement, rank-order behaviour, shared spectral area or local variation. The presentation also discusses how these choices relate to practical analytical contexts, from MIR fingerprint comparison to NIR applications dominated by broad, overlapping bands. Overall, the aim is to provide a clear methodological framework for selecting spectral comparison metrics according to the scientific question, the spectral domain and the intended decision.
| Presenting author | Roberto Sáez-Hernández |
|---|
