NIR Measurement of Moisture Content in Wood under Unstable Temperature Conditions. Part 2. Handling Temperature Fluctuations

Fluctuations in sample temperature cause peak shifts in near infrared (NIR) spectra of moist, solid wood samples, especially when the temperature varies around 0°C (the freezing point of water). These thermal effects cannot be ignored when NIR and Partial Least Squares (PLS) regression is used for determination of the moisture content of wood outside the laboratory. In this paper, a number of different approaches to the problem are investigated. The approaches may be divided into two different classes according to their basic strategy. One strategy, the soft model strategy, is to represent all relevant temperatures in the calibration set and then produce a global model or a set of local models based on raw or pretreated spectra. This strategy does not require knowledge of the structure of the thermal effects, but a large calibration set representing all relevant temperatures is needed. Three approaches based on this strategy were tested. The other strategy, the transformation strategy, is to develop the moisture content model at one temperature and transform spectra recorded at other temperatures to this temperature. If an effective transformation algorithm can be found, this strategy should require less calibration data. Four new approaches based on the transformation strategy were developed and tested. The three soft model approaches gave similar prediction errors (RMSEP) for unknown samples (8–9%, expressed as moisture ratio, i.e. the moisture content in percent of the dry weight). None of the approaches based on the transformation strategy gave smaller prediction errors than the soft model approaches, but two of them gave only slightly larger prediction errors (RMSEP) than the soft model approaches (9–10% moisture).