Robustness of multidimensional calibrations: application to spectral data

The calibration of devices providing continuous signals, such as spectra, poses particular problems of robustness. A sound understanding of this aspect is therefore necessary and this approach is based on a hierarchical questioning of the significance of the influence quantity, controllability and measurability. In the case of an influential variable, uncontrollable and yet measurable, it can be corrected by modifying the measured signal, or the model; its output can also be corrected. In the worst case, where the variable is influential, but neither controllable nor measurable, methods in order to improve the robustness of calibrations are necessary.