矿区土壤重金属Cd高光谱建模与模型适应性分析

Hyperspectral Inversion of Heavy Metal Cd in Mining Area and Its Response to Concentration Gradient

以粤北南岭某矿区土壤为研究对象,采用传统采样检测方法分析土壤重金属含量。借助ASD FieldSpec 4型便携式高光谱仪测量土壤光谱反射率,将土壤反射率经过卷积平滑后进行8种数据变换,分析9种光谱数据指标与镉(Cd)含量的相关性。采用偏最小二乘和随机森林方法,结合Pearson相关系数分别建立9种光谱数据指标的土壤重金属Cd含量的高光谱反演模型。并对其预测能力进行评价,实现了局地尺度上土壤重金属Cd含量反演,探究了不同Cd浓度区间对建模精度的影响。结果表明:通过不同的数据变换方式可以有效消除基线,去除背景干扰并提高光谱数据与Cd的相关性。其中一阶微分变换效果最为理想;偏最小二乘(Partial Least Squares Regression,PLSR)与随机森林(Random Forest Regression,RF)均能不同程度预测Cd含量,其中采用一阶微分结合随机森林(FD-RF)方法所建模型具备较高可靠性,R^(2)超过0.80;当样本浓度平均值变化超过40%时,建立的模型预测能力降低。该研究方法可以作为土壤重金属Cd检测的手段,研究结果可以为土壤重金属高光谱反演提供方法和理论支持。

The soil of a mining area in the north of Guangdong province was taken as the research object,and the traditional sampling method was used to analyze the content of heavy metals in the soil,the reflectance spectra of soil were measured by ASD FieldSpec 4.Eight kinds of data were transformed after smoothing,the correlation between nine spectral data indexes and Cd content was analyzed.Combined with Pearson's correlation coefficient,the hyperspectral retrieval models of soil heavy metal Cd content with nine spectral data indexes were established,and their prediction ability was evaluated.The retrieval of soil heavy metal Cd content on local scale was realized,and the influence of different Cd content intervals on modeling accuracy was explored.The results show that the baseline,background interference and the correlation between the spectral data and Cd can be effectively eliminated by different data transformation methods,first order differential transformation is the most ideal;both PLSR and Random forest can predict Cd content to different degrees.Among them,the model established by the First-order Differential combined with Random Forest(FD-RF)method has a high reliability,and the R^(2) exceeds 0.85.When the mean of the sample concentration changes by more than 40%,the predictive power of the model was reduced.This approach can be used as a method for Cd detection of heavy metals in soil,and the research results can provide method and theoretical support for hyperspectral inversion of heavy metals in soil.