施用生物炭后土壤有机碳的近红外光谱模型研究与应用

土壤有机碳是影响土壤肥力的最重要因素之一。生物炭由于具有高度芳香化碳结构和发达孔隙结构等特性，可以作为一种土壤改良剂，提高土壤有机碳含量，改善土壤物理结构，近些年成为农业环境领域研究的热点。分别采用传统方法和可见光近红外光谱（VIS-NIRS，400-2500 nm）技术对施加不同用量生物炭的土壤有机碳含量进行检测和对比分析，以期为含生物炭土壤的有机碳分析建立有效预测模型。通过比较不同样本选择方法（Kennard-Stone（KS），Random selection（RS）和Sample set partitioning based on joint x-y distances（SPXY））和光谱预处理方法（Savitzky-Golay平滑（SG）、倒数的对数log（1/R）、标准正态变量变换（SNV）、一阶导数（Der1）、二阶导数（Der2）和多元散射校正（MSC）），以3种模型（组合间隔偏最小二乘模型（Synergy Interval Partial Least Squares，siPLS），遗传算法-支持向量机模型（Genetic Algorithm-Support vector machine，GA-SVM）和随机森林模型（Random Forest，RF））来建立生物炭土壤有机碳预测模型。结果表明：（1）施加生物炭增加了土壤有机碳含量，增加幅度随生物炭添加量的提高呈增加趋势；（2）土壤反射率随土壤有机碳含量的增加而降低，在1410、1920和2200 nm光谱附近存在明显的吸收谷；（3）对比3种样本选择方法，KS方法所划分的样本集相对于RS方法和SPXY方法更适用于生物炭土壤有机碳模型的建立；（4）以SG+MSC预处理结合GA-SVM方法建立的模型精度最高，校正集的R_cal²和RMSECV值分别为0.9526和0.4839，验证集的R²_val和RMSEP值分别为0.8598和0.9987，RPD值为2.6017。该模型因具有精度高且模拟效果较好等优点，可用于含生物炭土壤的有机碳含量的科学预测。

Study on near-infrared spectroscopy model of soil organic carbon after biochar addition and its application

Soil organic carbon (SOC) is one of the most important factors affecting soil fertility. Due to the characteristics of highly aromatic carbon structure and developed pore structure, biochar can be used as a soil amendment to increase SOC content and improve soil physical structure, which has become a research hotspot in the fields of agriculture and environment in recent years. In this study, both traditional method and visible near-infrared spectroscopy (VIS-NIRS, 400-2500 nm) were used to detect SOC content in samples containing different amounts of biochar, in order to establish an effective prediction model for the analysis of organic carbon in soils containing biochar. An optimal prediction model was established for quantifying SOC content through three processes, including comparing different sample-selection methods (Kennard-Stone, Random selection, and SPXY), comparing various spectral pre-processing methods, and matching with three models. The pre-processing methods included Savitzky-Golay smoothing (SG), log(1/R), standard normal variate transformation (SNV), first derivative (Der1), second derivative (Der2), and multiplicative scatter correction (MSC). The three models applied in this study were Synergy Interval Partial Least Squares (siPLS), Genetic Algorithm-Support Vector Machine (GA-SVM), and Random Forests (RF). Results showed that: (1) SOC content was increased significantly by biochar addition and was affected by the amount of biochar. (2) Soil reflectance decreased with the SOC content increasing, indicated by obvious absorption valleys at the spectra nearby 1410, 1920, and 2200 nm. (3) Compared with the three sample selection methods, the sample set divided by KS method was more suitable for the SOC modeling process than those by RS and SPXY methods. (4) The model established by SG+MSC pretreatment combining with GA-SVM method had the highest accuracy, with R_cal²=0.9526 and RMSECV=0.4839 in the calibration set, and R²_val=0.8598, RMSEP=0.9987, and RPD=2.6017 in the validation set. The model can be used for scientific prediction of SOC in samples containing biochar due to its advantages of high precision and good simulation effects.