苜蓿人工草地高光谱遥感估产模型的研究

Research on the hyperspectral remote sensing estimation models for the fresh yield of alfalfa grassland

实地测量了10个苜蓿品种在不同生长时期的冠层高光谱数据,用以消除不同品种间光谱的差异性,并以多个波段的反射率、一阶导数和多种光谱吸收特征参数为光谱参数,运用多种单变量回归模型,对人工苜蓿草地的鲜草产量进行了估算。结果表明,复合、指数等非线性模型要优于线性模型,在同类型的光谱参数中,线性模型决定系数高的参数,其二次型、三次型多项式和复合、乘幂、指数等非线性回归模型的决定系数通常也比较高,且通常高于线性模型;诸多估算模型中,以747 nm处一阶导数为自变量的复合、指数2种形式的估算模型,其相关系数最高为r=0.852,均方根误差为0.466 kg/m2,相对误差为21.14%,其估算精度最高,可作为多个苜蓿品种统一使用的鲜草产量高光谱估算模型。

In the field,hyperspectral data and fresh yields in different growth periods of 10 alfalfa （Medicago sativa） varieties were collected and then the reflectance,first derivative spectrum and spectral absorption feature parameters were used as independent variables,to build univariate regression models for estimating the fresh yield of alfalfa.The nonlinear regression equation models,such as quadratic,cubic,compound,power and exponential were better than linear models.Among the spectral parameters of the same type,the parameters which had high coefficient of determinations of the linear model,always had higher coefficients of determination of quadratic,cubic,compound,power and exponential models.Among these estimation models,the compound and exponential estimation models based on the first derivative value at wavelength 747 nm was the most accurate for estimating the fresh yield of alfalfa; The correlation coefficient （r） was 0.852,the root mean square error （RMSE） was 0.466 kg/m2,and the relative error （RE） was 21.14％.The models can be used to estimate the fresh yield of the 10 alfalfa varieties.