Comparison of different uni- and multi-variate techniques for monitoring leaf water status as an indicator of water-deficit stress in wheat through spectroscopy

Abstract

Ten different wheat genotypes were studied for understanding their differential behaviour to different water-deficit stress levels. Hyperspectral data (350–2500 nm) and relative water content (RWC) of plants were measured at different stress level for identifying optimal spectral bands, indices and multivariate models to develop non-invasive phenotyping protocols. Evaluation of water sensitive existing spectral indices, proposed indices and band depth analysis at selected wavelengths was done with respect to RWC and prediction models were developed. The prediction models developed were efficient in predicting RWC with the R2 values ranging from 0.73 to 0.88 for spectral indices and 0.74–0.85 with continuum depth. Then, the ratio spectral indices (RSI) and normalised difference spectral indices (NDSI) were obtained in all possible combinations within 350–2500 nm and their correlations with RWC were quantified to identify the best indices. The best spectral indices for estimating RWC in wheat were RSI (R1391, R1830) and NDSI (R1391, R1830) with R2 of 0.86 and 0.81, respectively. Spectral reflectance data were also used to develop partial least squares regression (PLSR) followed by multiple linear regression (MLR), support vector machine regression (SVR), multivariate adaptive regression spline (MARS) and random forest (RF) model to calculate plant RWC. Among these multivariate models, PLSR was the best model for prediction of RWC (R2 and RMSE of 0.96 and 3.88%; 0.91 and 6.52% for calibration and validation, respectively). The methodology developed would help for its further use in high-throughput phenomics of different crops for drought stress.