Multiblock-Based Qualitative and Quantitative Spectral Calibration Analysis

In this article, an improved spectral calibration and statistical analysis approach is proposed. Having realized the multiplicity of underlying spectral characteristics and their different effects on the quality of interpretation and prediction, the major task lies in how to qualify and quantify the descriptor-quality relationship more meaningfully over different wavelength subspaces. The underlying spectral information can be explored more comprehensively by a spectral subspace separation and multiblock modeling strategy. Unlike the specific purpose of quality predictions over different spectral subspaces, systematic information in both descriptor and quality spaces is decomposed into different parts under the control of different between-set relationships. Closely related variations and irrelevant ones are discriminated and evaluated separately, where, in particular, the orthogonal variations, as important systematic information, are quantified in specific model parameters, even though they are not predictive. In this way, the approach well tracks the wavelength-varying underlying characteristics, takes full advantage of underlying variations in both X and Y spaces, and allows for a more meaningful interpretation of the between-set relationship. This article theoretically and experimentally illustrates the efficiency of the proposed method in qualitative and quantitative spectral calibration analysis