传统的以“点采样+实验室分析”为主的土壤重金属含量分析技术成本高、 效率低下, 而基于多光谱遥感的土壤重金属高精度定量反演中存在重金属含量影响因子的优化这一难题, 以陕西大西沟矿区这类山区地形条件下的金属矿区为例, 利用Landsat8/OLI多光谱卫星影像、 DEM数据以及外业土壤采样分析数据, 开展了矿区土壤重金属含量指示因子分析及定量反演研究。 首先, 考虑研究区地形地貌特点, 设计了沿研究区地形特征线及其两侧坡面均匀分布的样点分布方案, 采集了45个样本。 并对45个样本的混合样中的8种重金属含量进行了兴趣度分析, 根据含量超标程度及矿的类型选取了铜、 铅、 砷3种元素作为分析对象。 其次, 根据研究区土地利用现状及地形特点, 提出了以Landsat8/OLI影像B2至B7波段光谱反射率、 粘土矿物比(CMR)、 改进归一化水体指数(MNDWI)、 差异植被指数(DVI)等八种光谱指数、 以及反映研究区地形坡度和坡向三类因子作为反映土壤重金属含量空间分布特征的侯选因子。 进而, 对上述三类侯选因子与样本中3种金属含量进行了最小二乘相关性分析。 根据分析结果, 引入了基于估算误差最小准则的金属含量估算模型——基于规则的M5模型树的分段线性估算模型。 以上述三大类共17个指示因子作为模型的输入, 利用80%的土壤样本分析数据作为模型的训练数据, 经过M5模型树的构建、 平滑和树枝修剪过程, 建立了3种金属的反演模型实现了研究区中土壤中3种金属含量的估算。 同时, 基于均方根误差(RMSE)最小准则确定了以光谱因子为主的最利于反演的最佳指示因子集。 最后, 用随机选取的20%的检验样本对模型进行了反演精度分析, 验证了该模型对铜、 铅、 砷3种金属含量的反演精度比普通的线性模型分别提高了27.3%, 24.6%, 20.9%, 同时, 铜、 铅元素的可信度也有所提高。 利用上述模型的反演结果实现了3种金属含量的空间分布制图, 并将反演结果与1990年公布的国家土壤元素背景值进行了对比。 此外, 分析了研究区铜、 铅、 砷3种金属的空间分布规律, 并利用野外调查结果进行了验证。

The problem of low efficiency and higher cost exists in the traditional method mainly on “field-work point sampling then indoor experimental analysis”. Also the problem that how to choose the optimal factors indicating the content of heavy metals in soils is difficult to solve for the quantitative inversion of high precision using multispectral remote sensing technology. Using Landsat8/OLI satellite imagery, DEM data and soil samples data, the paper performed the analysis indicators of heavy metals in soil and the quantitative inversion of the content of heavy metals in soil in order to achieve an improved accuracy, taking a study case of a mountainous and forestry mining area called Daxigou mineral of Shaanxi in China. The work was as follows: A soil sampling scheme considering terrain and geomorphology characteristics was designed and evenly sampled in both sides along main topographic feature lines in the study area and 45 soil samples were acquired. Furthermore, a mixed samples from 45 samples were analyzed in laboratory so as to choose the most interested metals (i.e. Cu, Zn, As) as our focus according to both the degree of metals content bigger than that of national authoritative statistics and the type of mineral. Secondly, the paper suggested three types of factors including six spectral reflectivity from band two to seven of Landsat8/OLI imagery, and several spectral indices such as CMR, MNDWI, DVI, EVI etc., derived from Landsat8 image and also slope and aspect factors derived from DEM data were adopted to indicate the characteristics of the spatial distribution of the content of the three metals candidates considering land use and terrain circumstances in the study area. Subsequently, a correlation analysis of the content of three interested metals individually with six spectral reflectivity data, eight spectral indices and three terrain indicators was done using Least Squares principle. According to the consequence of the correlation analysis, the paper introduced the rule-based M5 model tree in the form of piecewise linear model which was used to estimate the content of Cu, Zn, As three metals in the principle of minimizing error rate. And an inversion model for the content of the three metals was constructed through the simulation, smoothing and pruning of the model tree with an input of all three types and 17 indicators mentioned above and 80% training samples. Also, a set of optimal indicators focusing on spectrum for the inversion were determined according to the principle of minimizing RMSE. Finally, the inversion results using 20% random samples were verified, showing that our suggested method achieved a decrease of RMSE value by 27.3%, 24.6%, 20.9%, and an improvement in confidence level for Cu and As, compared to that of the three interested metals using ordinary linear regression model. Also the thematic images showing the spatial distribution were mapped using the model. Then, the comparisons between the estimated value of the content of three metals and the background value published by Chinese government in 1990 were made. Furthermore, the statistical distribution rules of the three metals were concluded and verified using field survey results.