Calibration And Validation
Al = log (1/Tl) = c* el *I
Matter-of-fact reasons for the choice of certain wavelengths and coefficients in such a calibration equation can be given only in very rare cases, since the calibration of NIRS instruments is made very difficult by the complexity of the spectra. At almost every wavelength, more than one ingredient exhibits absorptions. Therefore, overlapping of absorption bands in the near-infrared spectral range occurs in almost all measurements. As a result, it is not possible in near-infrared spectroscopy to use calibration series of pure substances or reference materials for calibration.
The calibration development procedure is mainly based on statistical techniques, and the calibration equation is determined empirically. Therefore, the quality of this calibration function must be tested with new samples before it can be used in routine operation. Such a test with a new, independent random sample is called validation. The figure graphically depicts the model of calibration and validation. The calibration equation is always compiled for a certain basic entity. The calibration and validation samples are independent random samples from this basic entity.
The term "calibration" stands for both the calibration development and the calibration function.
Calibration and validation in NIR spectroscopy; calibration on the left, validation on the right.