Prediction model of rice crude protein content, amylose content and actual yield under high temperature stress based on hyper-spectral remote sensing

Crude protein and amylose constitute two main representative components of rice quality. The non-destructive, quick assessment of grain crude protein content (GCPC), grain amylose content (GAC), and actual yield (AC) are necessary for quality and yield diagnosis in rice production. The objectives of this study were to determine the effects of high temperature stress on rice GCPC, GAC and AC, to define the relationships of GCPC, GAC, and AC to ground-based canopy hyper-spectral reflectance and derivative parameters, and to establish quantitative models for real-time monitoring of rice GCPC, GAC, and AC using sensitive spectral parameters under high temperature stress. Two field warming experiments were performed in Nanjing in Jiangsu Province, China, to investigate the effects of high temperature ((treated for continuous 3 days from 9:00 am to 14:00 pm, average temperature was set at 35, 38 and 41 degrees C) and a control (CK)) at flowering stage in Liangyoupeijiu rice cultivar, using a free air temperature increase apparatus. Canopy hyper-spectral reflectance, GCPC, GAC, and AC were measured under high temperature treatments during different growth stages (flowering stage, grain-filling and ripening stages). The results showed that GCPC, AC (GAC) in Liangyoupeijiu were clearly reduced (increased) under high temperature stress in this study compared with the values of CK, and the reducing extent of GCPC and AC (the increasing extent of GAC) was increased with the increase of high temperature level. The hyper-spectral reflectance in different wavelength regions under high temperature stress was different. They increased in visible light region with the elevation of temperature, but reduced in near-infrared region. Among some selected spectral indices at three different growth stages used to estimate GCPC, GAC and AC, the optimum indices were difference vegetation index(810,450) and perpendicular vegetation index- Landsat multispectral scanner with high R-2 when regressed against GCPC, GAC and AC. Furthermore, GCPC, GAC and AC prediction based on flowering stages were preferred than that on grain-filling and ripening stage by much bigger correlation coefficients. The six regression models developed in this study showed the agreement between the predicted and observed values when testing independent data under high temperature stress. Thus, the selected key hyper-spectral parameters can be reliably used to estimate GCPC, GAC and AC in rice under different high temperature treatments.