GRAIN STARCH ESTIMATION USING HYPERSPECTRAL DATA AND ITS RELATIONSHIP WITH LEAF WATER CONTENTS FOR BROOMCORN MILLET (PANICUM MILIACEUM L.)

The analysis of grain starch content is one of the main indices to determine crop quality. The experiment was performed using two broomcorn millet varieties (Neimi 2 and Ningmi 14) in northern China, having 5 different sowing dates. The prediction models for grain starch fraction were constructed by integrating the spectral models and grain starch content with leaf water content using the intersection approach. We have identified 468, 630, 806, and 1488 nm as characteristic bands of the first derivative spectrum having a positive correlation. The correlation coefficient between the vegetation index of 608 nm and 1488 nm was significantly higher than other characteristic bands regarding leaf water contents. It is therefore identified that RDVI (1488, 806) could be used effectively to monitor leaf water content of broomcorn millet under different sowing date treatments, R-2 = 0.7075, RMSE = 0.073, RPD = 1.430 in the calibration set, R-2 = 0.8458, RMSE = 0.042, RPD = 2.546 in the validation set. In addition, the monitoring models of RVI (1488, 806), RDVI (1488, 806), and PVI (1488, 806) were superior to other models, and RPD was 0.020, 1.978, and 1.904, respectively. Overall, the RDVI (1488, 806) monitoring model, due to its stability and higher accuracy provided precise data on the grain starch fraction of broomcorn millet.