作 者 ：韩兆迎,朱西存,刘庆,房贤一,王卓远
期 刊 ：植物营养与肥料学报 2014年 20卷 6期 页码：1545-1552
Keywords:soil organic matter, envelope curve, hyper-spectrum, Yellow River Delta,
摘 要 ：【目的】土壤有机质(SOM)具有改良土壤结构、 促进团粒结构形成、 增加土壤疏松性、 改善土壤通气性和透水性以及促进植物生长发育的作用。传统测定土壤有机质的方法,虽然精度高,但是实时性差。本文通过对土壤高光谱数据进行变换和分析,筛选出与土壤有机质含量相关性高的敏感波长,构建能够实时、 快速反演黄河三角洲土壤有机质含量的数学统计模型。【方法】60个土壤样品采于黄河三角州。利用ASD Fieldspec3光谱仪,在室内环境下对黄河三角洲不同有机质含量的风干土壤样本进行了光谱测量,利用化学方法测定了土壤的有机质含量。在对土壤样品高光谱反射率进行去包络线处理的基础上,与土壤有机质含量进行相关分析,筛选敏感波长;运用主成分回归分析、 多元线性回归分析、 二次多项式逐步回归分析和支持向量机回归分析方法,分别建立了有机质含量的反演模型。【结果】确定了估测土壤有机质含量的敏感波长,建立了能够快速反演黄河三角洲土壤有机质含量的数学统计模型。从土壤光谱反射率曲线可以看出在1400 nm、 1900 nm和2200 nm等波段附近有十分明显的水分吸收谷。经对比相关性可以看出,去包络线的数据处理方法明显提高了光谱反射率与土壤有机质之间的相关性。1278 nm、 1307 nm、 1314 nm、 1322 nm、 1328 nm、 1334 nm、 1343 nm 7个相关性较高的波长作为估测土壤有机质含量的敏感波长。基于主成分回归分析、 多元线性回归分析、 二次多项式逐步回归分析和支持向量机回归分析方法,分别构建了反演有机质含量的模型。其中,二次多项式逐步回归模型校正集的决定系数达到了0.865,验证集的决定系数最大,达到了0.837,为黄河三角洲土壤有机质含量的最佳反演模型。【结论】去包络线的数据处理方法可提高光谱反射率与土壤有机质之间的相关性,确定的1278 nm、 1307 nm、 1314 nm、 1322 nm、 1328 nm、 1334 nm、 1343 nm 7个波长是估测黄河三角洲土壤有机质含量的敏感波长。由于二次多项式逐步回归模型校证集的决定系数最高、 均方根误差最小,其拟合效果最好。因此二次多项式逐步回归模型对反演黄河三角洲土壤有机质含量是最佳的。
Abstract:【Objectives】Soil organic matter (SOM) can improve soil structure, promote formation of granular structure, increase soil loose, improve soil air permeability and water permeability, and promote plant growth and development. Although the traditional method is acurate in determination of SOM, it cannot provide real-time measurement. The sensitive wavelengthshigh correlated with the SOM content are screened to build a quick statistic model to estimate the SOM content in Yellow River delta.【Methods】Sixty soil samples were collected in the Yellow River Delta. Mathematical statistics was used to build the inversion model by studying the relationship between the spectral reflectance and the SOM content. The SOM contents of soil samples were analyzed by a chemical method, and their hyperspectral reflectences were measured in an indoor dark room environment by ASD Fieldspec3 spectrometer. Sensitive wavelengths were screened according to their correction with the SOM content, after the handling method of continuum-removal. The principal component regression analysis, multiple linear regression analysis, quadratic polynomial stepwise regression analysis and the support vector machine (SVM) regression analysis methods were used to establish models for predicting the organic content.【Results】 From the soil spectral reflectance curve, around 1400 nm, 1900 nm and 2200 nm bands have obvious moisture absorption valley. It is obviously improved the correlation between the SOM content and spectral reflectance after processing method of continuum-removal by comparing correlation. The seven higher correlated wavelengths, 1278 nm, 1307 nm, 1314 nm, 1322 nm, 1328 nm, 1334 nm and 1343 nm, are considered as the sensitive wavelengths to estimate the SOM content.The quadratic polynomial stepwise regression model is prevent to be the best inversion model for SOM content prediction in the Yellow River delta with the determination coefficientof 0.865 and the the largest determination coefficient of the validation set, which reaches 0.837.【Conclusions】 The processing method of continuum-removal could improve the correlation between the SOM and spectral reflectance. The near infraredspectral of 1278 nm, 1307 nm, 1314 nm, 1314 nm, 1328 nm, 1334 nm and 1343 nm are the sensitive wavelengths for estimating the SOM content. The quadratic polynomial stepwise regression model has the best fitting effect, is appropriate for estimating the SOM content in Yellow River Delta.
全 文 ：