Nondestructive estimation of leaf nitrogen and chlorophyll contents in grapes using field hyper spectral data and support vector machines approach

Monitoring the content of chlorophyll and nitrogen in farms and gardens is a measure of vegetation healthy and the quantity and quality of the products. Usually, these parameters are measured by laboratory analysis, which requires cost, time and chemicals. In this study, the ability of field spectroscopy was evaluated as a fast, non-destructive and eco-friendly method for estimating these parameters in grape leaf. Therefore, the spectral curve was prepared in the range of 350-2500 nm from 180 grape leaf samples, which collected from 30 vineyards in the summer of 2017. Then the optimal spectral wavelengths and indices, in estimating these parameters, were determined by Partial Least Squares (PLS) regression. Finally, performance of the selected optimal variables was evaluated by multiple linear regression and support vector machines (SVM). The results of PLS showed that the wavelengths in vicinity of 2402, 946, 725 and 446 nm and the 690, 1370, 729, 438, and 366 nm, were considered as optimal variables in predicting the chlorophyll and nitrogen contents in grape leaves, respectively. Also, visible and red edge regions had the highest sensitivity to the explanation changes in these parameters. The results of modelling showed that in the best structures of SVM, chlorophyll and nitrogen were estimated at test stage with R2 about 0.91 and 0.72, respectively. Therefore, according to the acceptable obtained results, it is recommended to use field-based spectroscopy, spectral library formation and the introduction of optimal wavelengths to monitor other biochemical parameters in plant species as a new and efficient method.