Linear techniques to correct for temperature-induced spectral variation in multivariate calibration

The influence of external physical variation such as temperature fluctuations on near-infrared (NIR) spectra and their effect on the predictive power of calibration models such as PLS have been studied. Different methods to correct for the temperature effect by explicitly including the temperature in a calibration model have been tested. The results are compared to the implicit inclusion, which takes the temperature into account only through the calibration design. Two data sets are used, one well-designed data set measured in the laboratory and one industrial data set consisting of measurements for process samples. For both data sets, the explicit inclusion of the temperature in the calibration models did not result in an improvement of the prediction accuracy compared to implicit inclusion.