Near InfraRed Spectroscopy (NIRS) is a spectroscopic method that uses the interaction between the variable organic chemistry of wood and the NIR component of electromagnetic radiation (1000-2500nm) to affect the amount of light reflected back to the detector.
Wood is composed of cells which have thick walls made from organic substances, predominantly cellulose, hemicelluloses and lignin. The relative abundance of these constituents varies not only between individual cells, but even more between different layers in the cell wall. Wood density are a measure of the mass of these cell walls within a given volume of wood. Pulp yield is a measure of the mass of wall material remaining, relative to the original mass, after the pulping process is complete. For Kraft pulp yield, the pulping process is directed primarily at removing as much of the lignin as possible, while retaining as much of the cellulose.
Incident radiation (light) excites the various bonds in the wood sample, which absorb some of the energy. Different chemical bonds (e.g C=0, N-H) are affected by different parts of the electromagnetic spectrum and consequently the total amount of light absorbed at different wavelengths is a dependent upon the relative abundance of those bonds. The relative abundance is dependent upon the chemistry.
Thus the NIR spectrum obtained from any wood sample typically varies in small ways from one sample to another, because the chemistry is always a little different. It is this variability that is utilised in the development of NIR calibration models. Unlike many other spectroscopic methods, there is no characteristic peak(s) relating to cellulose or lignin, the height or area of which can be used to determine concentration.
NIR calibrations can perform poorly for a variety or reasons. Among these are the following.
1. A fundamental requirement of NIR is that the spectra obtained from a sample must be representative of the whole sample used to generate the reference chemical data.
Because NIR spectra lack any clear distinctive peaks that can be used in quantification, there is little information from the actual spectrum that be used to assess it’s quality. Wood can contain many defects (eg knots, compression wood, bark contaminants) that can contribute errors to spectra.
Likewise wood properties vary markedly within a tree. If a spectrum is obtained from a localised point within the tree, and then related to the chemistry obtained from a whole-tree sample, the calibration model will rely heavily on co-linearity between the properties of the wood from which the spectrum was obtained and the whole-tree values.
2. The NIR calibration relies on the quality of the chemical data.
If the reproducibility of the chemical data is poor (high standard errors) then the NIR calibration will perform less than optimally. For example it has been well demonstrated that NIR calibrations are capable of predicting Kraft pulp yield well. But if the predictions are poor, compared to available reference data, this may indicate that the reference data is the problem. NIR can in this way become a means of assessing the quality of the laboratory data. On several occasions we have found NIR predictions of KPY to correlate poorly with actual laboratory data. When we have bullt a calibration from these samples using the laboratory reference data the calibration is poor suggesting either point 1 above or poor lab data.
3. The property of interest is not amenable to NIR prediction.
This maybe the case if the property is not present in sufficient amounts to provide a strong enough influence on the light absorbed.