In this study, near-infrared hyperspectral imaging was applied to predict the water content of golden pothos (Epipremnum aureum) leaves, after which partial least squares regression (PLSR) analysis was performed to predict their averaged water content. The resulting predictive model was then applied to each single-pixel spectra in order to construct a water content image that could be used to evaluate the model's applicability to the single-pixel spectra through partial least squares score comparisons between the averaged spectra used for calibration and the single-pixel spectra. In the next phase, it was determined that a rebuilt PLSR predictive model based on the averaged spectra of an applicable pixel showed higher prediction accuracy than that of the original model. This study provides effective information about the limitations of prediction mapping and the optimization of pixel selections for better calibrations.