基于高光谱反射率的喀斯特地区土壤重金属锌元素含量反演

针对传统土壤重金属锌元素含量测定效率低下和喀斯特地区山高坡陡土壤样品采集难度大，亟需先进手段获取土壤重金属锌元素含量的要求，以喀斯特流域为研究区，利用电感耦合等离子质谱测定土壤样品的锌元素含量和分光光广度计采集土壤光谱数据。将所测定的原始光谱，经过连续统去除、一阶、二阶微分、倒数、倒数对数、倒数对数一阶、倒数对数二阶微分7种数学变换，基于高光谱吸收重金属元素的特征吸收带初步判断光谱特征变量，利用相关分析进一步筛选特征变量，运用逐步回归最终确定有效建模光谱变量。采用非线性和线性算法，揭示光谱敏感波段反射率与重金锌元素含量之间的映射关系，进行土壤重金属含量估测。结果表明：基于耦合的光谱特征变量甄选方式，锌元素的特征波段580，810，1 410，1 910，2 160，2 260，2 270，2 350，2 430 nm与铁氧化物、有机质、粘土矿物吸收带关联，表明一定程度上捕捉到喀斯特地区土壤重金属锌元素的光谱吸收特性；运用随机森林、支持向量机、偏最小二乘3种算法进行元素含量与光谱变量建模后，采用决定系数和均方根误差评价模型精度。从光谱变换形式和模型性能二个维度综合判断，基于二阶微分变换的随机森林算法准确度最高，为最佳估算模型。通过高光谱反射率估测重金属锌元素含量，实现了喀斯特地区土壤重金属锌元素含量的高效快速反演，为喀斯特地区重金属元素含量动态监测提供了可靠的技术支撑。

Prediction Soil Heavy Metal Zinc Based on Spectral Reflectance in Karst Area

In order to solve the problem of inefficiency in measuring heavy metal zinc contentand soil samples collection difficultly with traditional way in karst area, it is greatly essential to get zinc content in soil by effective measures. The institutional area is a typical Karst region, soil zinc content as well as reflectance spectral of soil data were collected by inductively coupled plasma mass and Spectrophotometer. The reflectance spectra of measurement were handed by these steps. Firstly, 7 kinds of mathematical transformations were used including continuum removed, first differential, second differential, reciprocal transformation, absorbance transformation, first differential of absorbance, and second differential of absorbance. Secondly, spectral characteristic variables were determined based on the characteristic absorption band of spectral absorption of heavy metals. And then, further spectral characteristic variables were selected by correlation analysis. Finally, stepwise regression was used to determine the effective modeling spectral bands. Mapping relationships between Spectral bands and heavy metal zinc content were revealed by linear and nonlinear estimation algorithm, and the results aim to measure the heavy metal zinc in soil. It shows that the characteristic bands of zinc are associated with iron oxide, organic matter and clay mineral absorption band. It’s focused on 580，810，1 410，1 910，2 160，2 260，2 270，2 350，2 430 nm, and these results reveal that the absorption characteristics of heavy metal zinc possible were captured in karst area. The models were funded by Random Forests, Support Vector Machines, Partial Least Squares Regression to precision evaluation by coefficient of determination and the root mean square error of prediction. The best estimation model was obtained from spectrum transformation and model performance. The algorithm of Random forests for second differential transformation has the highest accuracy and is chosen as the best model. The content of heavy metal zinc was estimated by spectral reflectance. It is a rapid, efficient method for indirect evaluation of zinc. It provides a technical support for the dynamic monitoring of heavy metal content in karst areas.