基于高光谱数据的盐渍化土壤含盐量预测模型研究

Prediction model of salt content in salinized soil based on hyperspectral data

对采自内蒙古自治区科尔沁右翼中旗巴彦淖尔苏木典型盐渍化土壤进行光谱反射率测定,分析不同盐渍化程度土壤的光谱特征;利用均方根、对数、对数倒数、倒数及一、二阶微分等多种原始光谱反射率变换形式的多元逐步线性回归分析方法,构建预测土壤含盐量的模型。结果表明,虽然供试土壤含盐量不同,但其光谱曲线在形态上保持一致,土壤含盐量越高其反射率曲线就越高。光谱反射率与土壤含盐量的正负相关性在经过数学变换后得到增强,尤其是在一、二阶微分变换后明显增强:土壤含盐量≤7g/kg的土壤样本(非盐土)的原始光谱对数一阶微分反射率与土壤含盐量相关性最高,在1490nm处相关系数最大为-0.5898;土壤含盐量>7g/kg的土壤样本(盐土)的原始光谱对数一阶微分反射率与土壤含盐量相关性最高,在727nm处相关系数最大为-0.5591。利用多元逐步线性回归建立的预测模型,非盐土以原始光谱的二阶微分模型为最优,R^(2)=0.7292;盐土以对数二阶微分模型为最优,R^(2)=0.8718。模型可用于盐碱土土壤含盐量的快速测定。

Spectral reflectance of typical salinized soil collected from Bayannur Sumu,Horqin Middle Banner of Inner Mongolia Autonomous Region was measured,and spectral characteristics of soil with different salinization degrees were analyzed.Original spectral reflectance was transformed into square root,logarithm,logarithmic reciprocal,reciprocal and its first and second order differential,and prediction model of soil salt content was established by using multiple stepwise linear regression analysis method.Results showed that,spectral curves of soil with different salt content were basically the same in morphological characteristics.The higher the salt content,the greater the reflectivity.After mathematical transformation,positive and negative correlation between spectral reflectance and soil salt content was enhanced,especially after first-order and second-order differential transformation,correlation was significantly enhanced.For soil samples with salt content less than 7 g/kg(non-saline soil),correlation between original spectral logarithmic first-order differential reflectance and salt content was the highest,and correlation coefficient was the highest at 1490 nm,which was -0.5898,while for soil samples with salt content greater than 7 g/kg(saline soil),correlation between original spectral logarithmic first-order differential reflectance and soil salt content was the highest,correlation coefficient was the highest at 727 nm,which was−0.5591.In prediction model established by multiple stepwise linear regression,the second order differential model of original spectrum was the best for non-saline soil,and the R^(2) was 0.7292.Logarithmic second order differential model was the best for saline soil,and the R^(2) was 0.8718.It could be used for rapid determination of salt content in saline-alkali soil.