利用高光谱参数反演水稻叶片类胡萝卜素含量

为了探讨快速、准确预测水稻(Oryza sativa)叶片类胡萝卜素(Car)含量的敏感光谱波段和光谱指数, 通过实施涉及不同年份、不同生态点、不同施氮水平和不同品种类型的4个田间试验, 于主要生育期同步测定了水稻顶部4张叶片的光谱反射率及Car含量, 系统分析了350-2 500 nm范围内任意两波段组合而成的比值(SR (λ1, λ2))、归一化(ND (λ1, λ2))及已报道的对Car敏感的光谱指数与水稻叶片Car含量间的定量关系。结果表明, 不同Car含量水平下水稻叶片光谱反射率存在着明显变化, 以绿光及红边波段对水稻叶片Car含量变化最为敏感。723 nm附近的波段与近红外波段的比值组合以及713 nm附近的波段与近红外波段的归一化组合可以较好地预测水稻叶片Car含量, 以SR (723, 770)和ND (770, 713)表现最好, 线性拟合R²分别达到0.897和0.898。基于独立的试验资料的检验表明, 预测值和实测值的拟合R²分别为0.856和0.858, 均方根差RMSE均为0.072, 平均相对误差RE分别为11.9%和12.0%, 表明SR (723,770)和ND (770, 713)可有效地估算水稻上部叶片的Car含量。

Estimating leaf carotenoid content with hyperspectral parameters in rice

Aims Our objectives were to detect the relationship between leaf carotenoid (Car) content and spectral reflectance characteristics and to recommend useful hyperspectral wavebands and hyperspectral indices for nondestructiveand quick estimation of Car content in rice (Oryza sativa).Methods Four field experiments with different nitrogen application rates and rice cultivars were conducted at different eco-sites over three years. Time-course measurements were taken on hyperspectral reflectance of 350–2 500 nm and Car content in four top leaves. We calculated the simple ratio spectral index (SR (λ1, λ2)) and normalized difference spectral index (ND (λ1, λ2)) with all combinations of two wavelengths (λ1 and λ2 nm) as well as other indices sensitive to Car, and analyzed the relationships between Car content to single wavelength reflectance and these spectral indices.Important findings Spectral reflectance varied with Car content, and the sensitive wavebands mostly occurred at green and red edge regions. The SR indices using reflectance around 723 nm combined with near infrared reflectance (NIR) or the ND indices using reflectance around 713 nm combined with NIR could be used to estimate leaf Car content in rice, among which the SR (723, 770) and ND (770, 713) have the best performance, with determination of coefficients (R²) of 0.897 and 0.898, respectively. Tests with an independent dataset showed that R² values between observed and predicted Car content with SR (723, 770) and ND (770, 713) were 0.856 and 0.858, with root meansquare error (RMSE) as 0.072, and relative error (RE) as 11.9% and 12.0%, respectively, which indicated that Car content in top leaves of rice could be predicted effectively with these two indices.