The influence of sample temperature variation on l-AA determination using Fourier transform infrared attenuated total reflectance terahertz (FTIR-ATR-THz) spectroscopy was investigated. Using the best local calibration models at 22, 31 and 40 °C, prediction results for a prediction sample set at any one temperature were excellent, with low RMSEP and high SDRpred values. However, bias and SEP were significantly increased when samples from different temperatures were used in these models
higher RMSEPs resulted. On the other hand, global calibration models based on combinations of different temperatures gave better prediction results, with lower RMSEP and higher SDRpred values for all prediction samples at 22, 31 and 40 °C. With these global calibration models the bias predictions were also significantly decreased
resulting in lower RMSEP values.