可见光-近红外、中红外光谱的土壤有机质组分反演

土壤有机质是土壤肥力的物质基础，其含量的高低是评价土壤肥力的重要标志。土壤有机质组分根据其溶解性可分为胡敏素(HM)、胡敏酸(HA)、富里酸(FA)，不同组分的肥力特性差异显著，因此，土壤有机质组分数据可更加全面、客观的反映土壤肥力状况。传统土壤土壤有机质及组分的测定工序繁杂，效率低下且时效性差，大量研究表明高光谱技术能有效提高土壤属性的检测效率并降低测试成本，但关于可见光-近红外、中红外光谱检测土壤有机质组分的报道鲜见。为了探索中红外光谱及可见光-近红外-中红外组合光谱对土壤有机质组分检测的可行性，并对比有机质单一光谱模型与有机质不同组分的组合光谱模型的预测精度，以南疆地区农田土壤为例，在阿克苏及和田地区共采集93个土样，进行有机质、胡敏素、胡敏酸、富里酸含量及光谱数据的测定。其次，利用可见-近红外(VNIR)、中红外(MIR)及其组合光谱(VNIR-MIR)三种光谱数据集，采用偏最小二乘(PLSR)、支持向量机(SVM)、随机森林(RF)三种建模方式对土壤有机质、胡敏素、胡敏酸、富里酸含量进行组合模型分析预测。结果表明：(1)土壤有机质及各组分均与光谱反射率有较好的相关性，土壤有机质及组分在MIR谱段的特征波段数量明显多于VNIR谱段。(2)有机质最优预测模型的模式为VNIR-MIR-RF，该模型的决定系数R2为0.90；胡敏素与胡敏酸最优预测模型的模式均为VNIR-RF模型，R2均为0.92；富里酸最优预测模型的模式为MIR-RF模型，R2为0.94。(3) 基于胡敏素、胡敏酸和富里酸的有机质组合光谱模型的预测精度明显高于有机质单一光谱模型，两种模型的R2分别为0.93和0.90。实现了土壤有机质组分的高效快速反演，且基于有机质组分的组合模型提高了土壤有机质预测精度，为南疆地区大尺度土壤肥力的鉴定与精准施肥提供重要的参考价值。

Inversion of Soil Organic Matter Fraction in Southern Xinjiang by Visible-Near-Infrared and Mid-Infrared Spectra

Soil organic matter is the material basis of soil fertility, and its fraction is an important indicator to evaluate soil fertility. Soil organic matter fractions can be divided into humin (HM), humic acid (HA) and fulvic acid (FA) according to their solubility. The fertility characteristics of different fractions are significantly different. Therefore, the data of soil organic matter fractions can reflect the status of soil fertility more comprehensively and objectively. The traditional determination of soil organic matter and its fractions is complex, inefficient and time-effective. Many studies show that hyperspectral technology can effectively improve the detection efficiency of soil properties and reduce the testing cost, but the reports on the detection of soil organic matter fractions by visible-near infrared and mid infrared spectroscopy are rare. 93 soil samples were collected and analyzed to acquire the content and spectral information of SOC, HM, HA in Aksu and Hetian, southern Xinjiang, and to explore further the feasibility of mid-infrared spectroscopy and visible near infrared-mid infrared combined spectroscopy in detecting soil organic matter fractions and to comparing the prediction accuracy of a single spectral model for organic matter with that of a combined spectral model for different soil organic matter fractions. Secondly, three kinds of spectral data sets of visible near-infrared (VNIR), mid-infrared (MIR) and their combined spectra (VNIR-MIR) were used to analyze and predict the contents of soil organic matter, HM, HA and FA by using three modeling methods of partial least squares (PLSR), support vector machine (SVM) and random forest (RF). The results show that: (1) soil organic matter and its fractions had a good correlation with spectral reflectance, and the number of characteristic bands of soil organic matter and its fractions in Mir was significantly more than that in VNIR. (2) The optimal prediction model of organic matter is VNIR-MIR-RF with R2 of 0.90; the optimal prediction model of HM and HA is VNIR-RF model with R2 of 0.92; the optimal prediction model of FA is VNIR-RF model with R2 of 0.94. (3) The prediction accuracy of the organic matter combination spectral model based on HM, HA and FA is significantly higher than that of the single spectral model. The R2 of the two models is 0.93 and 0.90, respectively. The results of this study realized the efficient and rapid inversion of soil organic matter fractions, and the combined model based on organic matter fractions improved the prediction accuracy of soil organic matter and provided important reference value for large-scale soil fertility identification and precision fertilization in southern Xinjiang.