
中图分类号 S153.621 文献标识码A 文章编号 1007-7731(2025)15-0089-05
DOI号 10.16377/j.cnki.issn1007-7731.2025.15.022
Influence of different spectral transformation forms on the accuracy of partial least squares estimation model of soil organic matter
ZENG Yuanwen FAN Wenwu
(Chongqing Geomatics and Remote Sensing Center, Chongqing 401147, China)
AbstractThisstudyused field-colectedsoil samplesas test subjectsto conduct experiments including soil organicmater (SOM) content determination,hyperspectral data acquisition,and preprocessing.Sixspectral transformationswereapplied to the preprocessd spectral data:absorption depth (Depth),firstderivativeof logreflectance (FD-lgR),second derivativeof log-reflectance (SD-lgR),secondderivativeof reflectance (SD-R),second derivative ofreciprocal reflectance (SD-1/R),andsecondderivativeof reciprocallog-reflectance (SD-1/lgR).Partial least squares regression (PLSR) models for SOM estimation were establishedunder diffrent spectral transformation forms to analyze thecorrelation between spectral transformationsand SOM content,as wellas their impacton model accuracy.Theresults showed thatall6transformations exhibited bands significantlycorrelated with SOMcontent,with FD-lgRhaving the highest numberofsignificantlycorrelatedbands (71).TheFD-lgRmodelachievedadetermination coefficient ( R2 )of 0.995,a root mean square error of calibration (RMSEC) of 0.O63,a cross-validation R2 of 0.775,and a relative percent difference (RPD)of 2.681,allof which were among the highest values acrossall transformations.The scater plot of predicted versus measured values indicated that theFD-lgR model's estimates were close to the actual values,with an R2 of 0.872. Overall, the regression model based on FD-lgR demonstrated high accuracy and good stability.These findings provide a reference for subsequent hyperspectral data preprocessing and estimation model construction for soil organic matter.
Keywordssoil organic matter; hyperspectral; spectral transformation; partial least squares regression
土壤有机质(Soilorganicmatter,SOM)是土壤的重要组成部分,其含量是评价土壤肥力的重要指标;也是农作物生长的重要养分之一,对作物生长有显著影响。土壤有机质常规调查采用现场取样加室内测试的方法,存在费时、费力和无法大面积铺开等问题,因此,为快速、准确和大范围地获取土壤有机质含量信息,必须寻找新的技术方法来满足现代精准农业的发展需求。光谱分析技术的发展,为上述问题的解决提供了新的路径。由于其速度快、成本低、无污染,以及可同时反演多种成分等特点,已成为替代化学检测的有效手段之一[1-2]。研究表明,土壤有机质在可见光波段和近红外波段展现出独特的光谱特性,其含量是影响土壤光谱特性的核心要素[3]。陈颂超等4研究发现,水稻土在可见光与近红外波段和中红外波段的光谱吸收特性与其有机质含量有一定的相关性。邬登巍等分析了不同母质和土地利用类型对土壤有机质含量光谱预测模型精度的影响,并讨论了该模型的适用性。
近年来,除了对土壤有机质本身的光谱特性进行研究外,还对土壤光谱处理方式、土壤粒径大小、土壤光谱数学变换形式和反演模型等对土壤有机质估算的影响进行了深入探究。刘效栋研究发现,就王壤有机质反演模型而言,偏最小二乘模型优于多元线性回归模型,其具有更好的精度和稳定性。郄欣等将光谱数据进行了4种变换,并分别构建了SVR估算模型,结果显示,基于倒数对数一阶微分这种变换形式下的模型精度最高。……