Predicting the Protein Content of Winter Wheat Using Hyper-Spectral Images.

The product of leaf area index and SPAD value (LAI × SPAD) represents total chlorophyll content of the plant canopy and was related to the protein content of wheat (Triticum aestivum L.) grains. The objectives of this study are (1) to develop new optical indices for LAI × SPAD and (2) to determine the optimum timing using hyper-spectral images. We made a field experiment for two cultivars of winter wheat (Yumechikara and Kitahonami) using a hyper-spectral camera (custom-made, Genesia) at Hokkaido Agricultural Research Center, Sapporo. Using the hyper spectral reflectance, we explored Normalized Differential Spectral Indices defined as NDSI [i, j] = (R_i − R_j) / ( R_i + R_j), where R_x is the reflectance at x nm, and suffixes i and j are wavelengths. As a result of the regression analysis between LAI × SPAD and NDSIs from any combination of two wavelengths, the coefficient of determination was highest with NDSI [780, 720] for Yumechikara (r² = 0.76) and for Kitahonami (r² = 0.83). To determine the best timing for the prediction of grain protein, we compared protein content and NDSI time series. As a result, the highest coefficient of determination was found on around 30 days post anthesis (DPA) for both the cultivars: Yumechikara (7 July, DPA 28, r² = 0.85) and Kitahonami (11 July, DPA 32, r² = 0.89). Canopies began to change color to yellow on around DPA 30. We found that NDSI determined using hyper-spectral images is available as an optical index to estimate LAI × SPAD. In addition, we determined the optimum timing to predict the protein content of grains using NDSI. However, wavelengths selected for NDSI were different for the two cultivars. Further studies are needed to validate the relationship between spectral reflectance and morphological traits of wheat crops.