基于高光谱遥感的棉花叶片叶绿素含量估算

Estimation on chlorophyll content of cotton based on optimized spectral index

为提高高光谱植被指数对棉花叶绿素含量的估算精度,以陕西省关中地区棉花花铃期叶片高光谱反射率为数据源,分析了13种植被指数与棉花叶片叶绿素相对含量(SPAD)的相关关系;同时采用降精细采样法,详细分析400~2 000nm波段范围内原始光谱反射率的任意两两波段组合而成的优化光谱指数RSI与SPAD值的定量关系,构建线性及非线性回归监测模型,并对模型进行验证。结果表明:1)所提取的13种植被指数中NIR/NIR与SPAD值的相关系数最大(r=0.914),并且基于NIR/NIR(R780/R740)构建的回归方程模型优于其他植被指数,其构建的二次曲线方程回归模型建模与验模R2分别为0.900和0.785,RMSE为4.762,RE为7.86%,为基于提取的12种植被指数构建SPAD值估算模型中最佳模型;2)优化后的比值光谱指数RSI(Ration spectral index)的敏感波段为500和563nm,RSI(500,563)与SPAD值的相关系数r=0.999,与棉花叶片SPAD含量在0.01水平下呈显著相关,其构建的二次曲线方程模型效果最优,建模和验模R2分别为0.912和1.000,RMSE为2.848,RE为4.38%。与提取的13种植被指数相比,基于RSI指数二次曲线回归模型为估算叶绿素含量的最佳模型,并且模型预测值和实测值之间的符合度较高R2=0.843,表明基于波段优化算法的优化光谱指数RSI能更好的预测棉花叶片叶绿素含量。

In order to improve the estimation accuracy of cotton chlorophyll content based on hyperspectral vegetation index,the spectral reflectance of cotton leaves in blooming stage in Guanzhong region was applied to analyze the relationship between 12 extracted vegetation index and chlorophyll content of cotton leaves. By using fine sampling method,a systematic analysis was undertaken on the quantitative relationship between chlorophyll content of cotton leaves and major hyper spectral index RSl （Ration spectral index）, which was composed of any two wavebands with original reflectance within the full spectral range of 400 - 2 000 rim. The results showed that= 1） The correlation coefficient between SPAD value and NIR/NIR was 0. 914, which is highest among the twelve extracted vegetation index. And the regression equation model based on NIR/NIR was better than model based other extracted vegetation index. The conic regression model was the best of all SPAD estimation models. The determination coefficients of training model and test model were 9. 900 and 0. 785, respectively, RMSE was 4. 762, and RE was 7.86%; 2） The optimized ratio of spectral index RSl of the sensitive wavelengths were 500 and 563 nm, respectively. The correlation coefficient between RSI （500,563） and SPAD value was 0. 999, and SPAD value of cotton leaves under the 0.01 level was significantly correlated. The conic regression model was the best of all SPAD estimation models. Its determination coefficients of training model and test model were 0. 912 and 1 respectively,RMSE was 2. 848,and RE was 4.38%. In conclusion,the extracted twelve vegetation index model based on quadratic equation had high degree of coincidence between the estimated and measured values. The analyses above further proved that the optimized spectral index RSI based on the wave bands combination could better predict the chlorophyll content of cotton leaves.