探讨组合模型在土壤重金属含量高光谱估算中的可行性, 以四川古蔺工矿废弃地复垦土壤为研究对象, 采集样品并测得重金属含量(Cd、Cr、Ni、As和Hg)和光谱信息; 对土壤光谱进行预处理, 探索响应波段; 利用PLS、ANN和RF构建单一估测模型; 利用熵值法进行较优模型组合。结果表明, 4种光谱预处理技术对光谱与重金属含量之间的相关性均有不同程度的提升; 单一预测模型MSC-RF模型效果最优, 可最大程度对重金属含量进行预测; 通过熵值法构建的组合预测模型较单一预测模型效果有所提升, 验证集R2最高达到0.91; 表明组合模型能够较大限度利用多种单个模型的样本信息, 减少单个模型中随机因素的影响, 增强模型的稳定性, 对矿业废弃地复垦土壤重金属含量的预测具有更好的发挥作用。

This paper was aiming to explore the feasibility of the combined model of hyperspectra on soil heavy metal content. Taking the reclaimed soil in Gulin, Sichuan Province as the research object, the samples were collected and the contents of heavy metals (Cd, Cr, Ni, As and Hg) and spectral information were measured. The soil spectrum was preprocessed to explore the response band; PLS, ANN and RF were used to construct a single estimation model, and entropy method was used for better model combination. The results showed that the correlation between heavy metal content and spectrum was improved by four kinds of spectral pretreatment techniques. The single prediction model MSC-RF had the best effect and could predict the heavy metal content to the maximum extent. Compared with the single prediction model, the combined prediction model constructed by the entropy value method has an improved effect, and the verification set R2 reaches a maximum of 0.91. It was showed that the combined model could make use of sample information of multiple single models to a large extent, reduce the influence of random factors in a single model, enhance the stability of the model, and play a better role in the prediction of heavy metal content of soil in abandoned mining land.