The complexity of near-infrared spectra
The pictures below show to the left an example of a simple single peak absorbance (Rhodamine B in water), where the concentration can be found simply by measuring the peak absorbance. To the right is a typical near-infrared spectrum (wheat) where the concentration of water and protein is very difficult to relate to a single peak in the spectrum.
Due to the broad and overlapping peaks found in near-infrared spectra, advanced numerical (software) methods like Chemometrics are typically used to extract parameters like concentrations from the spectral data. These methods are quite complex and typically builds on multi-variate statistical methods.
The overall process used in chemometrics is shown below:
The calibration step is where you build your model from a large set of NIR spectra measured on samples with different but known concentrations. A common method used in Principal Components Analysis (PCA) which takes the 100s of correlated spectral data (intensity versus wavelength) and reduces it to a small set of new independent variables – called Principal Components (PC).
In the validation step, you test how well your model can predict the correct concentrations of a known set of samples. The figure below shows a typical example of a model that provides a fairly linear relation between the known reference concentrations and the concentrations predicted by the chemometric model. Thus, we may use this model to predict concentrations based on near-infrared spectra measured on unknown samples.