ABSTRACT

This study aimed to investigate quantitative inversion of salt ions. Now, most research on hyperspectral characters in salinized soils focus on quantitative inversion of salt content, but few on ions. Original hyperspectral data were collected and calculated average as reflectance (R), the logarithm of the reciprocal of R (Log(1/R)) and continual removed reflectance (R_cr ) were also used. Correlations between salt content and ions were calculated, three ions were chosen as characteristic. SVR and PLSR methods were used to derive inversion models. Results showed both methods provided accurate measurement. For modeling precision, no matter which hyperspectral characters based on, inversion models derived with SVR were more accurate than PLSR. For test precision, inversion models derived with SVR were also more accurate. Inversion model of Cl^- based on R was the most accurate, followed by that based on R_cr , while the inversion model of K⁺ based on R_cr  was the most accurate. All the inversion models of Na⁺ based on R, Log(1/R) and R_cr  had good test precision. In conclusion, the accuracy of the quantitative inversion model with SVR was higher than PLSR, the best hyperspectral characters for inversion models were chosen, making a useful diagnostic tool for soil salinity.

摘要

这项研究旨在调查盐离子的定量反演。现在，大多数对盐渍化土壤中高光谱特征的研究都集中在盐含量的定量反演上，而很少涉及离子。收集原始高光谱数据和作为反射率（计算的平均值- [R ），的倒数的对数 - [R （日志（1 / [R ））和持续去除反射率（[R _CR ）也被使用。计算了盐含量与离子之间的相关性，选择了三个离子作为特征。SVR和PLSR方法用于推导反演模型。结果表明两种方法均可提供准确的测量结果。对于建模精度，无论基于哪个高光谱特征，使用SVR导出的反演模型都比PLSR更为准确。为了测试精度，使用SVR导出的反演模型也更加准确。的Cl反演模型^- 基于 ř 是最准确的，其次，基于 ř _CR ，而K的反演模型⁺ 基于 ř _CR  是最准确的。 基于Na ⁺的 所有Na ⁺反演模型R，Log（1 / R）和 R _cr  具有良好的测试精度。综上所述，SVR定量反演模型的准确性高于PLSR，为反演模型选择了最佳的高光谱特征，为土壤盐分化提供了有用的诊断工具。