Comparison of prediction power of three multivariate calibrations for estimation of leaf anthocyanin content with visible spectroscopy in Prunus cerasifera

The anthocyanin content in leaves can reveal valuable information about a plant’s physiological status and its responses to stress. Therefore, it is of great value to accurately and efficiently determine anthocyanin content in leaves. The selection of calibration method is a major factor which can influence the accuracy of measurement with visible and near infrared (NIR) spectroscopy. Three multivariate calibrations including principal component regression (PCR), partial least squares regression (PLSR), and back-propagation neural network (BPNN) were adopted for the development of determination models of leaf anthocyanin content using reflectance spectra data (450–600 nm) in Prunus cerasifera and then the performance of these models was compared for three multivariate calibrations. Certain principal components (PCs) and latent variables (LVs) were used as input for the back-propagation neural network (BPNN) model. The results showed that the best PCR and PLSR models were obtained by standard normal variate (SNV), and BPNN models outperformed both the PCR and PLSR models. The coefficient of determination (R(2)), the root mean square error of prediction (RMSE(p)), and the residual prediction deviation (RPD) values for the validation set were 0.920, 0.274, and 3.439, respectively, for the BPNN-PCs model, and 0.922, 0.270, and 3.489, respectively, for the BPNN-LVs model. Visible spectroscopy combined with BPNN was successfully applied to determine leaf anthocyanin content in P. cerasifera and the performance of the BPNN-LVs model was the best. The use of the BPNN-LVs model and visible spectroscopy showed significant potential for the nondestructive determination of leaf anthocyanin content in plants.