Wavelength selection for estimating biomass, LAI, and leaf nitrogen concentration in winter wheat of Gifu prefecture using in situ hyperspectral data

Abstract

Winter wheat (Triticum aestivum L.) is an important rotation crop in southern Gifu prefecture, but rotational use of dried paddy fields often reduces grain quality. Consequently, grain quality varies greatly throughout a single area. This study is intended to evaluate in situ canopy reflectance to predict the total dry matter (DM), leaf area index (LAI), and leaf nitrogen concentration (N_conc). In a field experiment including four levels of N application, hyperspectral reflectance data (400–950 nm) were obtained for winter wheat during five growth stages. Band selection applied using a normalized difference vegetation index (NDVI) formula yielded the best-fitted two-pair waveband. Several paired bands were also indicated, which had higher R² values of DM (R² > 0.8), LAI (R² > 0.4), and N_conc (R² > 0.5) than those in typical red and near-infrared (NIR) based NDVI(R² = 0.39, 0.27, and 0.01 respectively in DM, LAI, and N_conc). Partial least squares (PLS) regression analyses using the best two-pair NDVI predictions improved the predictive accuracy for DM (R² = 0.93), LAI (R² = 0.80), and N_conc (R² = 0.83). Furthermore, the model using first derivative reflectance (FDR) more precisely predicted DM (R² = 0.95) and N_conc (R² = 0.88) than that using reflectance spectra. These findings suggest that PLS can provide better prediction than conventional methods using a single waveband or band combination. Strong correlations were found in regions with shorter wavelengths: green(525–570 nm), red edge (720–780 nm), and NIR (930–950 nm) regions to DM and LAI; and blue (425–490 nm) and red regions (645–685 nm) to N_conc. These results suggest that important wavelength regions are green, red edge, and NIR regions for DM and LAI, and blue and red regions for N_conc.