基于高光谱特征参数荒漠草原生物量估测研究

Study on Biomass Estimation of Desert Steppe Based on Hyperspectral Characteristic

应用高光谱技术对荒漠草原草地生物量进行估测,分别采用原始光谱、一阶微分光谱和高光谱特征参数,分析荒漠草原草地冠层光谱数据与草地生物量相关关系,选择与草地生物量显著相关的高光谱特征参数,利用单变量线性与非线性、逐步回归和BP(Back Propagation)神经网络构建估测模型。试验结果表明:与草地生物量相关系数最大的为高光谱特征参数SDR/SDb,相关系数为0.792;通过高光谱特征参数Dy、Rg、Rr、SDy、SDR、Rg/Rr、(Rg-Rr)/(Rg+Rr)、SDR/SDb、(SDR-SDy)/(SDR+SDy)建立的BP神经网络回归模型最优,决定系数R2为0.766,均方根误差为40.19g/m2,相对误差为29.88%。因此,利用草地冠层高光谱特征参数可以估测草地生物量,能够对草地生物量进行定量分析,为草地精细化管理提供技术支持,对畜牧业可持续发展具有重要意义。

The hyperspectral technique was used to estimate grassland biomass of desert steppe.The relationship between grassland canopy spectral data of desert steppe and grassland biomass was analyzed by using the primitive spectrum,the first derivative spectra and hyperspectral characteristic parameters.Chose hyperspectral characteristic parameters which were associated with grassland biomass,and used univariate linear,nonlinear,stepwise regression and BP（Back Propagation） neural network to construct estimation model.The results show that the highest correlation coefficient with grassland biomass is hyperspectral characteristic parameter SDR/SDb,the correlation coefficient is 0.792.BP neural network established by hyperspectral characteristic parameters Dy,Rg,Rr,SDy,SDR,Rg/Rr,（Rg-Rr）/（Rg ＋ Rr）,SDR/SDb,（SDR-SDy）/（SDR ＋ SDy） is optimal,the coefficient R2 is 0.766,the root mean square error is 40.19 g/m2,and the relative error is 29.88%.Therefore we use hyperspectral characteristic parameters of grassland canopy to estimate grassland biomass,provide quantitative analysis for grassland biomass,and provide technical support for grassland fine management,which is of great significance to the sustainable development of animal husbandry.