Abstract: Grain protein content is an important index indicating wheat quality status, and non-destructive and quick assessments of grain protein content are necessary for cultural regulation and quality classification in wheat production. The objectives of this study were to determine the relationships of grain protein content to ground-based canopy hyper-spectral reflectance and spectral parameters, and to derive regression equations for predicting grain protein content in winter wheat (Triticum aestivum L.) with canopy hyper-spectral remote sensing, by four field experiments with different wheat cultivars and nitrogen levels across three growing seasons, and by time-course measurements on canopy hyperspectral reflectance, plant dry weight, nitrogen content and grain protein content during the experiment periods in 2003–2006. In experiment one and four two cultivars of Ningmai 9 and Yumai 34 was used as low and high protein types respectively. In experiment two, Yangmai 12 was added as medium protein type cultivar. In experiment three, the high protein type cultivar altered for Xuzhou 26, and medium type cultivar for Yangmai 10 and Huaimai 20. The results showed that the grain protein content at maturity in wheat increased with nitrogen rate promotion, and could be well estimated by plant nitrogen nutrition status such as leaf N content and leaf N accumulation at anthesis. The regression analyses between vegetation indices developed and leaf N nutrition indices such as leaf N content and leaf N accumulation indicated that several key spectral parameters could be accurately used to estimate the changes in leaf N status across different growth stages, nitrogen levels and growing seasons, e.g., REPle and mND705 could be used to leaf nitrogen content, and SDr/SDb and FD742 to leaf nitrogen accumulation. The total predicting models on grain protein content at maturity were constructed based on canopy hyper-spectral parameters at anthesis by linking the above two models with leaf N nutrition as intersection in wheat. Testing of the predicting models with independent datasets indicated that the spectral indices of REPle, mND705, SDr/SDb, and FD742 could be used to accurately estimate grain protein contents in wheat. It can be concluded that the grain protein content at maturity in wheat could be predicted directly by key vegetation indices at anthesis, with more reliable estimation from mND705.

Key words: Wheat, Nitrogen nutrition, Grain protein content, Hyper-spectral remote sensing, Prediction model