Determination of temperatures of aqueous-based samples directly from near infrared spectra

Abstract

The temperature sensitivity of underlying water absorption bands can lead to baseline artifacts or apparent spectral band shifts in near infrared spectra and can negatively impact multivariate calibration models used in quantitative analyses. To address this issue, efforts can be made to suppress the temperature-induced spectral variation or knowledge of the temperature can be used to adjust the calibration. To facilitate the latter approach, we explored the ability to estimate the aqueous temperature of the sample directly from the combination region of the near infrared spectrum. This temperature modeling strategy addresses applications in which it is difficult to obtain an accurate sample temperature with a conventional measurement probe. Temperature models were developed by use of partial least-squares regression combined with the discrete wavelet transform. Models were constructed from the 5000 to 4000 cm−1 region of near infrared spectra for pH 7.4 buffer solutions over the temperature range of 20.0–40.5℃. The long-term predictive ability of the models was assessed by use of 13 sets of prediction spectra collected over the course of 13 months, yielding values of the root mean square error of prediction ranging from 0.19 to 0.36℃. In addition, laboratory-prepared solutions of glucose, mixture solutions of glucose, lactate, urea in buffer, and bovine plasma were used to assess the predictive ability of the temperature models in increasingly complex matrixes. The effects of pH and buffer molarity were also studied. While increasing the complexity of the spectral background resulted in increases in root mean square error of prediction (0.33–1.01℃), retuning the models to incorporate the modified spectral backgrounds lowered the resulting root mean square error of prediction values to the range of 0.3℃. This work demonstrates the practical utility of spectral-based temperature measurements that employ the absorbance of the water baseline rather than the peak absorbance.