Sun Weichao, Liu Shuo, Wang Mengfei, Zhang Xia, Shang Kun, Liu Qingjie
Aerospace Information Research Institute, Chinese Academy of Sciences, No.9 Dengzhuang South Road, Haidian District, Beijing 100094, China.
Aerospace Information Research Institute, Chinese Academy of Sciences, No.9 Dengzhuang South Road, Haidian District, Beijing 100094, China.
Sci Total Environ. 2023 Feb 20;860:160511. doi: 10.1016/j.scitotenv.2022.160511. Epub 2022 Nov 25.
Hyperspectral remote sensing has the advantages to predict and map soil heavy metal concentration over conventional monitoring methods and multispectral remote sensing. In quantitative applications of hyperspectral remote sensing imagery, the contribution of hyperspectral bands is different, and abnormal prediction values resulted from incorrectly classified bare soil images are a major problem. In this study, a variable weighting method was proposed to weight the hyperspectral bands, and a probability threshold was used to improve the classification to mitigate the problem of abnormal prediction values. The variable weighting was conducted by using the absorption depths obtained by continuum removal. Soil samples were collected from a mining area in southwestern China. Hyperspectral remote sensing imagery was acquired by the Advanced Hyperspectral Imager (AHSI) abroad on Geofen-5 (GF-5) satellite. Genetic algorithm and partial least squares regression (PLSR) were adopted to calibrate prediction models. In prediction of soil copper (Cu) concentration, root mean square error (RMSE) and coefficient of determination (R) were 21.59 mg kg and 0.60 for the prediction using raw reflectance spectra, and the values were improved to 18.33 mg kg and 0.71 by using the weighted reflectance spectra. The developed prediction model was applied to the AHSI imagery to predict Cu concentration in bare soil areas. In prediction of Cu concentration using the AHSI imagery, negative prediction values were eliminated by using the bare soil image extracted by the improved classification. Based on the prediction, soil Cu concentration map was generated by kriging spatial interpolation. The result indicates that the proposed variable weighting method is effective and the problem of abnormal prediction values could be mitigated by using improved bare soil images. Further analysis indicates that some indices with proper thresholds also could be used to get improved bare soil images.
与传统监测方法和多光谱遥感相比，高光谱遥感在预测和绘制土壤重金属浓度方面具有优势。在高光谱遥感影像的定量应用中，高光谱波段的贡献各不相同，裸土图像分类错误导致的异常预测值是一个主要问题。本研究提出了一种可变加权方法对高光谱波段进行加权，并使用概率阈值改进分类以减轻异常预测值问题。可变加权通过使用连续统去除法获得的吸收深度来进行。土壤样本采集于中国西南部的一个矿区。高光谱遥感影像由国外搭载在高分五号（GF-5）卫星上的先进高光谱成像仪（AHSI）获取。采用遗传算法和偏最小二乘回归（PLSR）校准预测模型。在土壤铜（Cu）浓度预测中，使用原始反射光谱进行预测时，均方根误差（RMSE）和决定系数（R）分别为21.59 mg/kg和0.60，使用加权反射光谱时，这些值分别提高到18.33 mg/kg和0.71。将所建立的预测模型应用于AHSI影像，以预测裸土区域的铜浓度。在使用AHSI影像预测铜浓度时，通过改进分类提取的裸土图像消除了负预测值。基于该预测，通过克里金空间插值生成了土壤铜浓度图。结果表明，所提出的可变加权方法是有效的，使用改进的裸土图像可以减轻异常预测值问题。进一步分析表明，一些具有适当阈值的指数也可用于获得改进的裸土图像。
