高光谱遥感为冬小麦氮含量的实时估测提供了技术途径, 然而在实际探测过程中, 接收的信号不仅包含植株叶、 茎等器官内部发生多次散射后的光辐射, 也包含在叶片表面发生镜面反射而没有进入器官内部的光辐射, 原理上只有前者可反映植株的生化组分信息, 因此目前常用的反演算法存在较大不确定性。 拟采用增加偏振测量的方式, 区分与上述两种情形对应的非偏振光和部分偏振光, 通过构建相应的反射率表征因子, 以评估剔除部分偏振反射分量对植株氮含量估算的影响。 实验获取了冬小麦拔节、 挑旗、 开花、 灌浆四个典型生长期共计48组偏振高光谱与氮含量测量样本, 分析后表明, 剔除偏振反射后, 反射率光谱与氮含量的相关性在可见光波段有较明显的提升, 而常用的多个植被指数对氮含量的估算精度有小幅提升, 且不同生长期对应的最优植被指数不同。 上述结果证明了通过测量叶表偏振反射而提升冬小麦氮含量高光谱估算方法的有效性与稳定性, 为提升植被生化组分遥感反演的精度提供了参考。

Hyperspectral remote sensing provides an effective way for real-time prediction of plant nitrogen content (PNC) in winter wheat plants. In hyperspectral detection, energy received by the sensor is divided into unpolarized light, which comes from multiple scattering inside the plant, and partially polarized light, which is from the foliar surface, while the latter does not contain nitrogen content information. This paper aims to evaluate the influence of this part of the surface polarization reflection on the PNC estimation. The canopy bidirectional reflectance factor (BRF) in nadir direction of 48 plots in periods of jointing, flagging, flowering and grouting of winter wheat were obtained, and the polarization component was assembled in front of the spectrometer fiber. The polarized reflectance (pBRF) is obtained, and the diffused reflectance factor (dBRF), which partially removes the surface reflection, is obtained by removing the pBRF from the BRF. Using spectral regression and vegetation index (VI) methods, the results of BRF, dBRF, and the existing methods considering removing surface reflection, were compared, to prove the effectiveness and stability of the polarization method. Evaluate the correlation between PNC and BRF & dBRF spectrum; in the spectral regression method, interval partial least squares regression (iPLSR) was used for PNC estimation. The method considering first-order derivative BRF (derBRF) was also compared. For the VI method, the PNC-VI models were established by using 7 VIs. The existed modified VI (mVI) models were also compared for analysis of the advantages and stability of the polarization-dBRF method. Finally, the non-negligibility of polarization reflection, the accurate estimation of surface reflection and the main error sources of the experiment were discussed and analyzed. After the removal of the polarized reflection, the correlation between the reflectance spectrum and PNC is significant in the visible band. The correlation coefficient increased from 0.68 to 0.72 in the blue band and slightly increased in the other spectral regions. In the spectral regression method, the root means square error RMSE of the predicted-measured PNC and dBRF spectrum reduced from 0.30% to 0.23%, indicating 23%’s error reduction; the estimation result is better than derBRF. The method demonstrates the effectiveness of the polarization method. In the vegetation index method, the accuracy of the PNC estimation model of the 7 VIs after polarization removal is slightly improved, and the result is better than the mVI method, which proves the stability of the polarization method. The ND680 (NDVI), ND705 and OSAVI indices yielded better PNC estimation in flowering and grouting periods, with the modeling relative RMSE (RRMSE) within 11%; SR705 and NDNI performed the best in jointing and flagging periods, with the modeling RRMSE within 13%. This study provides a reference for improving the accuracy of remote sensing retrieval of vegetation components.