Prediction of nitrogen status based on agronomic and hyperspectral data at different growth stages of winter wheat

Abstract

Nitrogen management is an essential parameter for monitoring wheat plant growth, yield, and grain quality. Advance control management in wheat production generally requires non-destructive and rapid nitrogen determination. This experiment aimed to examine the relationship between spectral data and nitrogen contents in winter wheat (Triticum aestivum L.) at different growth periods, and develop a regression equation to predict nitrogen status. In this study, quantitative correlations between nitrogen content and canopy hyperspectral reflectance were examined under different nitrogen doses. This study was carried out in 2017 and 2018 at the Daxing experimental research base of the National Water Saving Irrigation Engineering Research Center in Beijing, China. The prediction accuracy of sensitive band 539 nm was maximum at the jointing-booting period with the maximum coefficient of determination(R²) lowest RMSE (Root mean square error) and RE% (Relative error %) (0.546, 0.424, 12.234). The prediction model accuracy of band 697 nm was the highest with maximum R2 and lowest RMSE and RE% (0.609, 0.406, 16.033) at the booting- heading period. The prediction model of band 700nm has the maximum R² and lowest RMSE and RE% (0.875, 0.489, 22.445) value at the heading-maturity period. The prediction accuracy of spectral index NDV1canste (normalized difference vegetation index caste) showed the highest prediction accuracy with the maximum R² and lowest RMSE and RE% (0.741, 0.411, 13.013) at the jointing-booting period, mND705 (modified normalized difference 705) showed the highest accuracy with the lowest RMSE and RE% (0.741, 0.41, 13.013) at the booting-heading period. At heading-maturity, NDVI2 showed the highest accuracy with R², RMSE and RE % value (0.824, 0.356, 16.376) in 2018. At the whole growth period, ND705 (normalized difference 705) performed the best accuracy with maximum R², RMSE, and RE% (0.90, 0.389, 16.453). In conclusion, these hyperspectral prediction models are the best for predicting nitrogen in winter wheat at various growth periods.

Keywords: Hyperspectral data; Nitrogen contents; Prediction model; Inter wheat

http://dx.doi.org/10.19045/bspab.2023.120112