基于水稻冠层光谱特征构建粳型水稻籽粒蛋白质含量预测模型

水稻籽粒蛋白质含量是评价稻米品质的主要指标之一。本文以不同施氮水平下的4年田间试验为基础，系统分析了水稻成熟籽粒蛋白质含量与不同时期冠层反射光谱的相关性。结果显示，籽粒蛋白质含量与可见光波段(460~710 nm) 反射率呈负相关，与近红外波段(760~1 220 nm) 反射率呈正相关，其中孕穗期冠层单波段反射率与成熟籽粒蛋白质含量的相关性最高，在16个波段中以760 nm波段反射率与籽粒蛋白质含量的拟合效果最好，复相关系数达0.795。进一步分析比值植被指数、差值植被指数、归一化植被指数和红边参数等光谱参数与成熟籽粒蛋白质含量的相关性，运用线性逐步回归分析方法对相关拟合较好的16个参数进行筛选，建立了水稻成熟籽粒蛋白质含量（GPC）监测模型，GPC=－0.15× DVI(1 500, 950) + 3.00。利用不同年份不同品种及不同施氮水平下的观测数据对模型进行检验，预测值和实测值的精确度为0.56~0.86，准确度为0.85~1.18，根均方差（RMSE）为3.51%~19.91%，表明模型预测值与实测值之间符合度较高，对水稻成熟籽粒蛋白质含量具有较好的预测性。

Estimating Grain Protein Content with Canopy Spectral Reflectance in Rice

Grain protein content is a key index for evaluating rice quality. We investigated the quantitative relationships between grain protein content and canopy reflectance spectra at different growth stages in rice (Oryza sativa L.) on the basis of the data from the field experiments involving different cultivars and nitrogen levels in four years. Experiment 1 and 2 were both conducted with one cultivar, Wuxiangjing 9, and five N application levels of 0, 75, 150, 225, and 300 kg ha^-1 in 2002 and 2003. Experiment 3 included three cultivars (Wuxiangjing 9, Huajing 2, Nipponbare) with four N application levels of 0, 105, 210, and 315 kg ha^-1 in 2004. Experiment 4 was designed with two cultivars (Wuxiangjing 14, 27123), and four N application levels of 0, 90, 270, and 420 kg ha^-1 in 2005. Canopy spectral reflectance (460–1 650 nm) date at jointing, booting, heading, filling, and ripening stages of rice in the different field experiments were measured with MSR-16 multi-spectral radiometer, and corresponding leaf area index and grain protein content were also determined. Then the relationships of grain protein content to canopy reflectance of single band and all two-band combinations were analyzed. The results showed that there were significant negative correlation between grain protein content and canopy spectral reflectance at 460–710 nm and positive correlation at 760–1 220 nm after jointing, with best performance from the relationship at 760 nm and booting stage. The relationships of grain protein content to the ratio, differential and normalized difference vegetation indices of all bands and red edge parameters were also analyzed, then fifteen parameters were selected. Through step regression analysis on 16 better spectral parameters, the differential vegetation index of R1500 and R950 was found to be the best parameter for predicting grain protein content (GPC) in rice. The derived equation, GPC =0.15 × DVI (1 500, 950) + 3, was tested with the observed data from the other independent experiments. The estimation precision was 0.56–0.86, estimation accuracy was 0.85–1.18, and RMSE was 3.51%–19.9%, indicating a good fit between the predicted and observed values of grain protein content. It is concluded that the present spectral index model is feasible and useful for estimating grain protein content in rice with different cultivars and nitrogen levels.