代希君, 张艳丽, 彭 杰, 罗华平, 向红英. 土壤水溶性盐基离子的高光谱反演模型及验证[J]. 农业工程学报, 2015, 31(22): 139-145. DOI: 10.11975/j.issn.1002-6819.2015.22.019
    引用本文: 代希君, 张艳丽, 彭 杰, 罗华平, 向红英. 土壤水溶性盐基离子的高光谱反演模型及验证[J]. 农业工程学报, 2015, 31(22): 139-145. DOI: 10.11975/j.issn.1002-6819.2015.22.019
    Dai Xijun, Zhang Yanli, Peng Jie, Luo Huaping, Xiang Hongying. Prediction and validation of water-soluble salt ions content using hyperspectral data[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2015, 31(22): 139-145. DOI: 10.11975/j.issn.1002-6819.2015.22.019
    Citation: Dai Xijun, Zhang Yanli, Peng Jie, Luo Huaping, Xiang Hongying. Prediction and validation of water-soluble salt ions content using hyperspectral data[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2015, 31(22): 139-145. DOI: 10.11975/j.issn.1002-6819.2015.22.019

    土壤水溶性盐基离子的高光谱反演模型及验证

    Prediction and validation of water-soluble salt ions content using hyperspectral data

    • 摘要: 土壤水溶性盐基离子是诊断土壤盐渍化类型与盐渍化程度的重要依据,利用光谱技术快速获取土壤水溶性盐基离子含量数据,可为土壤盐渍化类型与盐渍化程度的诊断提供新的技术和手段。该研究通过采集新疆5个不同地区399个土样的反射率与水溶性盐基离子数据并进行31种光谱预处理方法,分析了不同水溶性盐基离子(HCO3-、Cl-、SO42-、K+、Na+、Ca2+、Mg2+)与高光谱反射率之间的相关性;采用K-S(Kennard-Stone)方法挑选出299个样品,针对每种离子使用偏最小二乘回归(partial least squares regression,PLSR)分别建立32个高光谱定量反演模型,优选最佳反演模型,并单独使用100个样品对估测模型进行检验。结果表明:不同离子的最佳反演模型所使用的预处理方法存在差异,其中仅有Cl-和Ca2+、SO42-和Mg2+所使用的预处理方法相同,其他离子则不同;不同离子的反演精度也不同,HCO3-和Ca2+构建的模型相对分析误差(relative percent deviation,RPD)分别为2.67、2.57,模型具有很好的预测能力。Cl-、SO42-和Mg2+所构建的模型RPD分别为2.05、2.10和2.14,表明这三者建立的模型具有较好的预测能力。K+建立的模型RPD仅为1.11,不能对样品进行预测。Na+构建的模型RPD为1.83,表明该离子所建模型只能对样品进行粗略估测。研究结果为探究水溶性盐基离子的高光谱反演增添了新的内容,为土壤盐渍化监测的深入和推进提供了新的思路和方法。

       

      Abstract: Abstract: In soil, series of water-soluble salt ions can be used for identifying the types and degrees of soil salinization. Applying spectrum technology is a rapid way to detect the content of water-soluble salt ions in soil. To a certain extent, it could provide a new technique or channel for diagnosing the types and degrees of soil salinization. In this study, total of 399 soil samples were collected from Wensu county (105 samples), Baicheng county (79 samples), Awati county (60 samples), Xinhe county (48 samples) and Hetian county (107 samples) in southern Xinjiang province at a depth of 0-20 cm. The spectral reflectance and content of water-soluble salt ions (HCO3-, Cl-, SO42-, K+, Na+, Ca2+, Mg2+) were determined on all the soil samples and a number of salt ion testing methods were applied in this research. In order to find the most suitable spectrum pre-processing method, thirty one spectrum processing methods were conducted in the Unscrambler 9.7 software (a professional processing product for spectrum data), and through this way, the modeling accuracy may be promoted. The correlation between the content of water-soluble salt ions and the spectral reflectance was analyzed. Correlation analysis results on soil reflectance and different water-soluble salt ions showed that the correlation of K+ was the worst, and it was verified with the worst inversion accuracy in the validation process. Before being modeled, the single or combination pre-processing method was used as soil reflectance data. The original features masked showed up so that the correlation and the inversion accuracy were improved. Using K-S (Kennard-Stone) method, the total samples were divided into two groups with 299 samples for modeling and 100 samples for validating. Aiming at every water-soluble salt ion, thirty two corresponding quantitative inversion models were built through applying Partial Least Squares Regression (PLSR) method. Hereafter, the most suitable inversion model was identified among all the 32 models based on the predicting accuracy of the 100 validation samples. The research results demonstrated that the correlation curves of different water-soluble salt ions with their corresponding spectrum reflectance varied significantly. At the same time, the selected spectrum pro-processing methods for the specified water-soluble salt ions showed obviously differences. However, same processing methods were used for some of the salt ions. For example, the pre-processing method for Cl- and Ca2+ were the same, as well as for SO42+ and Mg2+. In addition, the methods applied for processing the spectrum data of other water soluble salt ions were different according to the predicting accuracy, such as HCO3-, K+ and Na+. The predicting accuracies of different water soluble salt ions were demonstrated differently. In terms of modeling process, the coefficients of determination termed as the R2 for HCO3- and Ca2+ were 0.859, 0.848, respectively. For the index of relative percent deviation termed as RPD for HCO3- and Ca2+ were 2.67, 2.57, respectively, which showed relatively good predicting abilities. Meanwhile, the R2 of Cl-, SO42+ and Mg2+ were 0.773, 0.781 and 0.761. At the same time, the RPD for these three salt ions were 2.10, 2.14 and 2.05, which displayed that the predicting abilities for Cl-, SO42+ and Mg2+ based on the spectrum technique were relatively better than K+, Na+ . As far as K+, the R2 and RPD for modeling in predicting K+ were 0.181 and 1.11, which showed the disabilities for predicting the content K+. Some results demonstrated roughly predicting abilities. For example, the R2 and RPD for Na+ were 0.702 and 1.83, respectively. The results and methods of this research expanded new applicable fields of spectrum technique in exploring the content of water soluble salt ions in soil. Consequently, this research provided new perspective and method for monitoring soil salinization.

       

    /

    返回文章
    返回