Near-infrared spectroscopy is a type of molecular vibration spectroscopy. Temperature variations in the sample cause changes in molecular vibrations and intermolecular forces, which in turn lead to absorption and spectral peak changes, affecting the prediction accuracy of minor components such as blood glucose. To address the impact of temperature perturbation on spectral detection and modeling analysis, a temperature perturbation discrimination method based on aquaphotomics and two-trace two-dimensional correlation spectroscopy (2T2D-COS) was proposed. The 2T2D-COS analysis was applied to diffuse reflectance spectra of simulated solutions under temperature perturbation and varying glucose concentrations. Spectral features induced by changes in temperature and glucose concentration were successfully extracted, revealing distinct water spectral patterns under different perturbations. Quantitative analysis shows that a temperature change of 0.1°C is equivalent to a glucose concentration change of 45 mg/dL in terms of intensity. A temperature perturbation anomaly spectral discrimination model was further established based on raw spectra, water spectral features, and 2T2D-COS asynchronous spectra. The accuracy rates of the models are 75%, 83.33%, and 95.83%, respectively. After removing anomalous samples, the root mean square error of glucose concentration prediction is reduced by