Abstract: Grain protein content (GPC) is a major index for evaluating the quality of winter wheat. To assess the influences of calibrated methods on the detection of GPC in winter wheat by using near infrared spectroscopy, original spectra reflectance was firstly pre-processed with S-G smoothing, baseline correction, and multiple scattering correction. The successive projectionalgorithm (SPA) method was used to extract the important spectral bands of grain GPC in winter wheat. Multivariate statistical methods of partial leastsquares regression (PLSR), principalcomponent regression (PCR), support vector machine (SVM), and multivariable linear regression (MLR) were used to establish a spectral prediction model of GPC, and the model accuracy was evaluated. The following wavelengths extracted by the SPA method were highly related to the GPC: 1801, 1010, 1109, 2219, 2239, 871, 1361, 2284, 1925, 1849 and 1456 nm. Wecompared the performance of models established with different multivariate methods. The results showed that the model of GPC constructed with SVM+SPA outperformed other calibrated models withR²=0.9760 and 0.9581, RMSE=0.2481 and 0.3587 for calibrating model and validated model, respectively. Our results indicated that the combination of SPA and SVM method might be a fast and non-destructive one to measure the grain protein of winter wheat.

Key words: soil microbial biomass nitrogen, rice, soil microbial biomass carbon, fertilizer management, soil microbial quotient