关键词: 干旱区, 土壤有机质, 分数阶微分, 机器学习

Abstract: In order to investigate the response mechanism between spectral reflectance and soil surface organic matter content based on fractional differential preprocessing, this study used soil canopy spectra of Anningqu Town, Urumqi County, Xinjiang as the data source. The GL fractional differential method was used to preprocess the hyperspectral data with orders ranging from 0 to 2.0 (with an interval of 0.2). The ratio spectral index, normalized spectral index, and difference spectral index based on fractional differential preprocessing were calculated using arbitrary band combination algorithms. The competitive adaptive reweighted sampling (CARS) algorithm was used to screen sensitive bands and spectral indices for soil organic matter content, combined with three machine learning algorithms (ANN, KNN, and SVM), a soil organic matter content estimation model based on fractional differential and machine learning methods was constructed, and model validation was performed. The results showed that there was a significant correlation between the spectral indices based on 0-2.0 orders and soil organic matter content. The correlation coefficients (r) of NDVI and RVI based on the original data and 0.2 order preprocessing exceeded 0.80. The Knearest neighbor algorithm based on the 0.2 order NDVI index exhibited the best ability to simulate soil organic matter content, with a coefficient of determination (R2) of 0.73, a root mean square error (RMSE) of 2.11 g/kg, and a relative analysis error (RPD) of 2.23, respectively. The study can provide theoretical support for remote sensing technology, achieve accurate monitoring and evaluation of soil fertility, and promote the development of smart agriculture.

Key words: arid area, soil organic matter, fractional differentiation, machine learning