Spectral reflectance variability from soil physicochemical properties in oil contaminated soils

详细信息	查看全文 | 推荐本文 |

• 作者：Somsubhra Chakraborty^e ; ¹ ; David C. Weindorf^a ; ^{DWeindorf@agcenter.lsu.edu} ; Yuanda Zhu^a ; ² ; Bin Li^b ; ³ ; Cristine L.S. Morgan^c ; Yufeng Ge^c ; John Galbraith^d
• 关键词：DRS ; diffuse reflectance spectroscopy ; LDA ; linear discriminant analysis ; LIBSVM ; library for support vector machines ; MLR ; multiple linear regression ; MR ; misclassification rate ; NIR ; near-infrared ; NIRS ; near-infrared spectrometry ; PC ; principal compone
• 刊名：Geoderma
• 出版年：2012
• 期刊代码：28_00167061
• 类别：ear
• 出版时间：May, 2012
• 卷：177-178
• 期：Complete
• 页码：80-89
• 文件大小：1376 K

摘要

Oil spills occur across large landscapes in a variety of soils. Visible and near-infrared (VisNIR, 350-2500 nm) diffuse reflectance spectroscopy (DRS) is a rapid, cost-effective sensing method that has shown potential for characterizing petroleum contaminated soils. This study used DRS to measure reflectance patterns of 68 samples made by mixing samples from two soils with different clay content, three levels of organic carbon, three petroleum types and three or more levels of contamination per type. Both first derivative of reflectance and discrete wavelet transformations were used to preprocess the spectra. Three clustering analyses (linear discriminant analysis, support vector machines, and random forest) and three multivariate regression methods (stepwise multiple linear regression, MLR; partial least squares regression, PLSR; and penalized spline) were used for pattern recognition and to develop the petroleum predictive models. Principal component analysis (PCA) was applied for qualitative VisNIR discrimination of variable soil types, organic carbon levels, petroleum types, and concentration levels. Soil types were separated with 100%accuracy and levels of organic carbon were separated with 96%accuracy by linear discriminant analysis using the first nine principal components. The support vector machine produced 82%classification accuracy for organic carbon levels by repeated random splitting of the whole dataset. However, spectral absorptions for each petroleum hydrocarbon overlapped with each other and could not be separated with any clustering scheme when contaminations were mixed. Wavelet-based MLR performed best for predicting petroleum amount with the highest residual prediction deviation (RPD) of 3.97. While using the first derivative of reflectance spectra, penalized spline regression performed better (RPD = 3.3) than PLSR (RPD = 2.5) model. Specific calibrations considering additional soil physicochemical variability and integrating wavelet-penalized spline are expected to produce useful spectral libraries for petroleum contaminated soils.