基于野外实测高光谱数据的干旱区耕作土壤含水量反演研究

Inversion models of soil water content using hyperspectral measurements in fields of the arid region farmland

利用2011年3月野外实地采集的不同含水量土壤的高光谱数据,研究了南疆地区耕作土壤草甸土含水量与高光谱反射率之间的定量关系,构建了一元线性回归与多元逐步回归的土壤含水量预测模型。结果表明,土壤含水量在380～1080 nm波段与反射率呈负相关关系;反射率经倒数(1/R)、对数(logR)、一阶微分(R′)变换后可提高其与含水量的相关性;以50个建模样本所建立模型的相关系数均达到极显著水平,所有模型通过对37验证样本进行预测,比较决定系数、均方根误差、相对误差后,表明多元逐步回归模型的预测能力要优于一元线性回归模型,从所有模型中优选出以698、702、703、746、747 nm波段反射率倒数(1/R)建立的多元逐步回归模型为最优模型,该模型实测值与预测值之间的R2为0.9199,RMSE为1.6026,RE为0.6517,可用于基于野外高光谱数据的土壤含水量的估测。

The quantitative relationships between soil water contents and hyperspectral reflectances in the southern Xinjiang farmland were studied based on hyperspectral measurements in fields in march of 2011.Prediction equations of soil water content were established using simple linear regression and stepwise multiple regression methods.Soil water contents were negatively related with spectral reflectances（SC） and its various transformations such as reciprocal of SC（1/SC）,logSC,first order differential of SC.Correlation coefficients of prediction equations based on 50 soil samples were significant at p0.01.An independent validation set of 37 soils showed that the stepwise multiple regression equation in which reciprocal of SC at 698nm,702nm,703nm,746nm and 747nm were defined as independent factors is the best model for fitting to the data of soil water content suggesting that the regression developed can accurately assess soil water content using hyperspectral measurements in fields.