土壤中的重金属含量较少, 难以在光谱曲线上表现出明显的特征, 现有的土壤重金属反演实验多是通过统计的方法寻找重金属的敏感波段, 不能准确解释土壤重金属的反演机理, 难以建立土壤重金属反演的普适性模型, 通过分析铁锰氧化物、 有机质、 粘土矿物在土壤光谱曲线上的吸收特征, 深入研究了土壤重金属对可见光近红外光谱的影响, 分析了褐土中的重金属反演机理。 以徐州试验田为例, 共采集80个土壤样本。 首先, 利用ASD地物光谱仪测定土壤样本的光谱反射率, 并采用电感耦合等离子体质谱仪检测土壤样品中的Cr, Cd, Cu, Pb和Zn的含量。 然后, 土壤光谱经过包络线去除处理, 与重金属相关的吸收峰在480, 1 780和2 200 nm附近, 所显现的吸收峰主要受土壤中的铁锰氧化物、 有机质、 粘土矿物的影响。 在吸收峰位置提取了光谱吸收特征的四个参数： Depth480, Depth1 780, Depth2 200和Area2 200, 分析了它们随五种重金属含量变化的增减趋势, 发现四个参数数值与五种重金属含量有很强的相关性。 分析单个变量反演重金属发现, 参数Depth480反演Cr和Pb的效果较好, 参数Area2 200, Depth1 780反演Cd, Cu和Zn的效果比较好。 同时使用四个光谱吸收特征参数, 利用最小二乘法、 岭回归法、 支持向量回归法求取回归系数, 建立的五种重金属含量的反演模型比使用单变量建立的反演模型预测能力强且稳定, 五种重金属Cr, Cd, Cu, Pb和Zn反演效果最好的验证集决定系数分别是0.71, 0.84, 0.92, 0.80, 0.89。 结果表明, 在此研究区域Cr和Pb容易被铁锰氧化物吸附, 而Cd, Cu和Zn更容易被有机质、 粘土矿物吸附。 此研究为探究土壤光谱特征与土壤重金属含量之间的关系提供了参考。

Heavy metals are scarce in soil, and it is difficult to identify their obvious characteristics in the soil spectrum. The previous soil heavy metal estimation methods have mostly applied statistical methods to find the characteristic bands, which cannot accurately explain the retrieval mechanism. It is therefore difficult to establish a universal model for soil heavy metal estimation. In order to investigate the influence of soil heavy metals in visible and near-infrared spectroscopy and analyze the retrieval mechanism of soil heavy metals, it is necessary to study the absorption characteristics of iron/manganese oxides, organic matter, clay minerals, etc. In this study, 80 soil samples were collected from the experimental field at Xuzhou, China. The spectra of the soil samples were measured with an Analytical Spectral Devices (ASD) field spectrometer. The soil heavy metal contents (Cr, Cd, Cu, Pb, and Zn) were determined by inductively coupled plasma-mass spectrometry. The soil spectra were processed by continuum removal. The absorption peaks related to heavy metals were around 480, 1 780, and 2 200 nm, which can be mainly attributed to iron/manganese oxides, organic matter, and clay minerals in the soil. The four spectral absorption characteristic parameters of Depth480, Depth1 780, Depth2 200, and Area2 200 were extracted at the positions of the absorption peaks. The variation trends of the parameters, along with the contents of the five heavy metals, were then analyzed. It was found that the four parameters were strongly correlated with the contents of the five heavy metals. Using a single variable to estimate the heavy metals, it was found that Depth480 had a higher estimation accuracy for Cr and Pb, and Area2 200 and Depth1 780 had a higher estimation accuracy for Cd, Cu, and Zn. The four spectral absorption characteristic parameters were implemented as independent variables, and the regression coefficients were obtained by ordinary least squares, ridge regression, and support vector regression. The heavy metal estimation model using the four spectral absorption characteristic parameters was stronger and more stable than those using only a single parameter. The best R2p (determination coefficient of prediction) values of the estimation models (Cr, Cd, Cu, Pb, and Zn) were 0.71, 0.84, 0.92, 0.80, and 0.89 respectively. The results suggest that Cr and Pb are easily adsorbed by iron/manganese oxides, while Cd, Cu, and Zn are more easily adsorbed by organic matter and clay minerals in this study area. The results of this study will provide a reference for researchers exploring the relationship between soil spectral characteristics and heavy metals.