光谱变换方法对黑土有机质反演模型的影响

Effect of Spectral Transformation Methods on the Inversion Model of Organic Matter in Black Soil

摘要: 土壤有机质高光谱反演是评估黑土肥力的重要手段，其精度受光谱预处理方法显著影响。本研究旨在系统评价不同光谱数学变换对黑土有机质反演模型精度与稳定性的影响。以黑龙江省拜泉县为研究区，获取350~2500 nm黑土高光谱数据。经剔除噪声波段、S-G平滑和重采样等预处理后，分析倒数、对数、倒数对数、一阶微分及其组合变换等8种数学变换的效果。结合敏感波段筛选与偏最小二乘回归(PLSR)建模，采用均方根误差(RMSE)、决定系数(R²)和相对分析误差(RPD)评估模型表现。结果表明，1) 不同数学变换显著影响敏感波段位置与数量；2) 倒数对数变换效果最优；3) 一阶微分及其组合变换在训练集表现优异，但测试集RPD均低于2，泛化能力不足；4) 多元散射校正、倒数、对数及原始光谱变换的模型测试集精度均未达理想水平(RPD < 2)。研究证实，倒数对数变换能有效提升黑土有机质高光谱反演模型的精度与稳定性，为土壤养分快速监测及精准农业管理提供了可靠技术支撑。

Abstract: Hyperspectral inversion of soil organic matter is an important tool for assessing the fertility of black soil, and its accuracy is significantly affected by spectral preprocessing methods. The aim of this study is to systematically evaluate the effects of different spectral mathematical transformations on the accuracy and stability of the black soil organic matter inversion model. Baiquan County of Heilongjiang Province was used as the study area to obtain 350~2500 nm black soil hyperspectral data. After the pre-processing of noise band removal, S-G smoothing and resampling, the effects of eight mathematical transformations, including inverse, logarithmic, inverse logarithmic, first-order differential and their combination transformations, were analyzed. Combining sensitive band screening with partial least squares regression (PLSR) modeling, the model performance is evaluated using root mean square error (RMSE), coefficient of determination (R²) and relative analysis error (RPD). The results show that 1) different mathematical transformations significantly affect the location and number of sensitive bands; 2) the inverse logarithmic transformation is the most effective; 3) the first-order differential and its combined transformations perform well in the training set, but the RPD of the test set is lower than 2, which is insufficient for the generalization ability; and 4) the accuracy of the test set of the model with multivariate scattering correction, the inverse logarithmic, logarithmic, and the primitive spectral transformations is not up to the desirable level (RPD < 2). The study confirms that the inverse logarithmic transformation can effectively improve the accuracy and stability of the hyperspectral inversion model of black soil organic matter, which provides a reliable technical support for the rapid monitoring of soil nutrients and the management of precision agriculture.