Department of Soil Science and Soil Protection, Faculty of Agrobiology, Food and Natural Resources, Czech University of Life Sciences Prague, Kamycka 129, Suchdol, Prague 16500, Czech Republic.
Department of Soil Science and Soil Protection, Faculty of Agrobiology, Food and Natural Resources, Czech University of Life Sciences Prague, Kamycka 129, Suchdol, Prague 16500, Czech Republic.
Sci Total Environ. 2023 May 10;872:161996. doi: 10.1016/j.scitotenv.2023.161996. Epub 2023 Feb 10.
Toxic elements released due to mining activities are of the most important environmental concerns, characterised not only by their concentration, but also by their distribution among different chemical species, known as speciation. These are conventionally determined using chemical analysis and sequential extraction, which are expensive and time-demanding. In this study, the possibility of using visible-near-infrared-shortwave infrared (VNIR-SWIR) reflectance spectroscopy was investigated as an alternative technique to quantify the contents of cobalt (Co) and nickel (Ni) in soil samples collected from Sarcheshmeh copper mine waste dump surface, in Iran. As a novel approach, the capability of VNIR-SWIR spectroscopy was also investigated in speciation of those elements. Three machine learning (ML) techniques (i.e., extreme gradient boosting (EGB), random forest (RF) and support vector regression (SVR)) were used to make relationships between soil spectral responses and Co and Ni contents of the samples. For all ML algorithms, the best prediction accuracies were obtained by the models developed on the first derivative (FD) spectra (for Co: RMSE values of 7.82, 8.03 and 9.22 mg·kg, and for Ni: RMSE values of 9.88, 10.32 and 11.02 mg·kg, using EGB, RF and SVR, respectively). Spatial variability maps of elements showed relatively similar patterns between observed and predicted values. Correlation and ML (EGB, RF, SVR)-based methods revealed that the most important wavelengths for Co and Ni prediction were those related to iron oxides/hydroxides and clay minerals, as two main soil properties responsible for controlling their speciation. This study demonstrated that the EGB technique was successful at indirect quantification and spatial variability mapping of Co and Ni on the mine waste dump surface. In addition, it provided an inspiration for implementation of the VNIR-SWIR reflectance spectroscopy as a potentially fast and cost-effective method for speciation studies of toxic elements, especially in heterogeneous soil environments.
采矿活动释放的有毒元素是最重要的环境问题之一，这些元素不仅具有浓度特征，而且还具有不同化学物质（称为形态）之间的分布特征。这些通常使用化学分析和连续提取来确定，这些方法既昂贵又耗时。在这项研究中，研究了可见-近红外-短波红外（VNIR-SWIR）反射光谱作为替代技术来定量测量伊朗 Sarcheshmeh 铜矿废堆表面土壤样品中钴（Co）和镍（Ni）含量的可能性。作为一种新方法，还研究了 VNIR-SWIR 光谱在这些元素形态中的能力。使用三种机器学习（ML）技术（即，极端梯度增强（EGB）、随机森林（RF）和支持向量回归（SVR））来建立土壤光谱响应与样品中 Co 和 Ni 含量之间的关系。对于所有 ML 算法，在一阶导数（FD）光谱上建立的模型都获得了最佳的预测精度（对于 Co：EGB、RF 和 SVR 的 RMSE 值分别为 7.82、8.03 和 9.22mg·kg；对于 Ni：RMSE 值分别为 9.88、10.32 和 11.02mg·kg）。元素的空间变异图显示了观测值和预测值之间相对相似的模式。相关性和基于 ML（EGB、RF、SVR）的方法表明，对于 Co 和 Ni 预测最重要的波长与铁氧化物/氢氧化物和粘土矿物有关，这是控制它们形态的两个主要土壤特性。本研究表明，EGB 技术成功地实现了对矿山废物堆表面 Co 和 Ni 的间接定量和空间变异性映射。此外，它为 VNIR-SWIR 反射光谱作为一种潜在的快速、经济有效的方法在异质土壤环境中进行有毒元素形态研究提供了启示。
