工业区土壤高光谱特性及其重金属含量预测研究——以上海市闵行区为例

Soil Hyperspectral Characteristics and Prediction of Heavy Metal Content in Soil of Industrial Area: A Case Study of Minhang District in Shanghai City

土壤的高光谱能够反映土壤信息,利用光谱数据能够快速高效地预测土壤重金属含量。以上海闵行工业区土壤为研究对象,分析研究区土壤光谱曲线所包含的土壤信息,通过光谱反射数据多种低阶微分变换形式以增强光谱特征,并通过相关分析获取土壤铜(Cu)、铬(Cr)、锰(Mn)、铅(Pb)、锌(Zn)含量的特征波段位置,构建土壤重金属高光谱反演模型。结果表明,工业区土壤重金属Cu、Cr、Mn、Pb、Zn存在明显积累,含量平均值依次为29.38、106.32、583.14、34.13、111.15 mg·kg-1,均超过上海市土壤环境背景值,且表现出复合污染和同源性的特点;土壤光谱曲线除受铁氧化物(409.55 nm)、有机质(825～900 nm)、水(1 400～1 425 nm)的影响存在光谱曲线吸收带外,走势均随波长增加而上升;最优特征波段均出现在近红外波段范围内,Cu为1 440.21 nm,Cr为834.44 nm,Mn为984.98 nm,Pb为1 440.21 nm,Zn为1 642.50 nm;利用特征波段光谱反射率的低阶微分变换形式作为建模因子建立土壤重金属含量的高光谱反演模型,发现均方根二阶微分(RMSTD)变换、倒数一阶微分(RTFD)变换、倒数对数二阶微分(ATSD)变换相较于原始一阶微分(FD)、二阶微分(SD)变换更能提高预测精度,多元逐步回归线性模型(SMLR)建模效果优于偏最小二乘回归(PLSR)模型,SMLR建模R2均大于0.65,验证R2除Mn外(R2=0.400)均大于0.5。基于SMLR建立的高光谱反演模型具有较高的预测精度、稳定性与可靠性,对工业区土壤重金属含量进行高光谱反演建模具有可行性。

Soil information can be reflected by soil hyperspectral spectrum,and the spectral data can be used to predict heavy metal content in soil.Urban soil of industrial area in Minhang District,Shanghai city was selected for research,and soil information contained in the soil spectral curve was analyzed.Soil spectral features were enhanced by various low-order differential transformations of soil spectral reflectance.The correlation analysis between various transformations of soil spectral reflectance and the content of copper(Cu),chromium(Cr),manganese(Mn),lead(Pb)and zinc(Zn)in soil was performed to obtain sensitive wavebands.An inversion model was developed for estimating heavy metal contents in soil.The results showed that the average contents of Cu,Cr,Mn,Pb and Zn in the soil were 29.38,106.3,583.1,34.13,and 111.2 mg·kg-1 respectively,and all the measured values of heavy metal contents exceeded the background levels of soil environment in Shanghai.There was obvious accumulation of heavy metals in the soil,characterized by compound pollution and pollution homology.Except for spectral absorption at 410,825~900,and 1 400~1 425 nm which were influenced by iron oxides,organic matter and water,the soil spectral curve tended to rise with increasing wavelength.The optimal feature band for soil Cu,Cr,Mn,Pb and Zn were at 1 440,834,985,1 440,and 1 643 nm,respectively in the near infrared range.Hyperspectral inversion models of soil heavy metals were established by low-order differential transformations of the characteristic band spectral reflectance.Compared with the first and second derivative of spectral reflectance,the prediction accuracy was improved by root mean square twice derivative,reciprocal transformation first derivative and absorbance transformation second derivative of the spectral reflectance.Compared with partial least-squares regression model,the stepwise multiple linear regression model performed better,with higher coefficient of determination(R2)for validation and model set.Except for soil Mn(R2=0.400),R2 for validation set was greater than 0.5,and R2 for model set was greater than 0.65.The inversion model based on stepwise multiple linear regression provided relatively high prediction accuracy,stability and reliability.These results indicate that it is promising to predict heavy metal content in the soil of industrial areas,using the newly developed spectral inversion model.