NIR Calibrations are based on mathematical models, based on spectral and reference data, and these data sources changes with time due to the following facts:
natural biological substances changes by season, weather, origin (continents, country), pollution, evolution of species, genetically changes, related especially to: Food & Feed
product production processes change by improvements, process parameters, timing, physical properties, environment, changed raw material variations, different suppliers, related especially to: Pharmacy, Chemistry, Process Analytical Technology (PAT), Quality by Design (QbD)
sampling process changes sample selection, see Theory of Sampling (TOS)
sample preparation changes chopping, grinding, milling, mixing (homogeneous), sieving, numerous effects of variable particle sizes, heating/freezing temperature program (ramp, hold), wet, dry, pressure and density, thickness, aging and contamination of samples between NIR and Lab measurement, air-tight transport cell, weight or volume, see NIR Sample Preparation
sample measurement changes measurement cell cleaning, container, glasses, petri glasses, cuvettes, plastic coverage, auto sampler adjustments and sampling plan, positioning, measurement area, fixed vs. moving spot, sample temperature, spectral resolution, apodization method (FT-NIR), number of scans, measurement repeats, averaging with/out outliers
reference method changes different method types, different Labs, refinements
SOP changes QA/QC procedures
instrument / spectrometer changes drifts by temperature and aging of electronic components, aging and defilement (dirt) of reference substances, wavelength accuracy, signal to noise ratio
new NIR data is collected continuously and should be used to extend the calibration, fill the matrix gaps, to increase robustness, and in some cases the older data can be faded out.
Because of all these changes, NIR Spectroscopy requires extensive application calibration and validation on an ongoing basis.
It’s like the weather forecast models, everything is changing and so the models need to be adjusted. Thankfully for NIR the period is longer than for the weather. But there is an interval, that means the models can not be held frozen and constant if the measurement results should be reliable. NIR require speriodic
