典型农田原状土壤剖面有机碳组分与结构的精细定量表征

It is of great significance to study the fractions and chemical structure of soil organic carbon (SOC) along soil profiles under the global climate change to evaluate the transformation, accumulation and cycle of deeper SOC caused by human actions. However, it is difficult to obtain the continuous distribution of SOC fractions along soil profiles. Moreover, the traditional measuring methods of SOC and its fractions are tedious and time-consuming, and they struggle to meet the rapid monitoring of modern agriculture and precise management. Based on the above requirements and scientific issues, this project intends to select four typical zonal soils in China, such as black soils, Fluvo-aquic soils, paddy soils and red soils, as research subjects. First, intact soil profiles (0–100 cm depth) under different farmland management practices are collected in the four soil types. Vis-NIR imaging spectroscopy (380–1000 nm) is used to obtain the hyperspectral data and image data of these soil profiles. Meanwhile, the SOC fractions and chemical structure are measured by the traditional grouping methods and solid 13C-NMR method, respectively. Then, the effects of different feature extraction methods and multivariate modeling methods on the prediction accuracy of SOC fractions are investigated to select the optimal prediction models for different SOC fractions. Finally, the continuous distribution of SOC fractions and chemical structure in the whole soil profile is quantitatively characterized and visually mapped at a fine spatial resolution. The main objective of this study is to quantify the vertical distribution patterns of SOC fractions and their structures under different management practices and soil types, which will provide a scientific basis for the further understanding the mechanism of SOC transformation and accumulation along soil profiles.

全球气候变化背景下研究土壤剖面有机碳（SOC）组成和结构对于评价人为措施引起的深层SOC转化和累积具有重要意义。然而目前研究中很难获取土壤剖面连续的SOC组分分布，且传统测量SOC组分方法步骤繁琐、耗时费工，难以满足现代农业快速监测、精确管理的需求。基于以上科学问题，项目拟选择东北黑土、华北潮土、太湖水稻土、南方红壤四大典型土壤为研究对象，采集不同管理措施下1 m深原状土壤剖面，通过Vis-NIR高光谱成像技术（380–1000 nm）获得土壤剖面高光谱和图像数据。分层测量SOC组分和化学结构（13C-NMR方法），探索不同特征变量提取和多变量建模方法对SOC组分预测精度的影响，筛选不同SOC组分的最优预测模型，最终反演整段原状土壤剖面SOC组分含量及其结构的连续分布，并可视化制图。精细定量表征不同管理措施下不同土壤类型SOC组分及其结构的垂直分布模式，明确剖面SOC转化和累积的作用机制。

全球气候变化背景下研究土壤剖面有机碳（SOC）组成和化学结构具有重要意义。然而目前研究中很难获取土壤剖面连续的SOC组分分布，且传统测量SOC组分方法步骤繁琐、耗时费工。项目采集黑土、潮土、水稻土等典型土壤类型1 m深原状剖面，通过Vis-NIR高光谱成像技术（380–1000 nm）获得土壤剖面高光谱图像，并分析剖面SOC及活性组分（包括水溶性碳DOC、易氧化碳ROC、微生物量碳MBC等）和SOC化学结构特征，探索不同特征变量提取和多变量建模方法对SOC组分预测精度的影响，筛选不同SOC组分的最优预测模型，最终反演整段原状土壤剖面不同碳组分含量的空间分布，并可视化制图。研究结果表明：（1）基于验证集数据，4种非线性模型（ANN、Cubist、GPR和SVMR）预测土壤碳组分含量的模拟性能明显明优于线性PLSR模型；全谱FS-SVMR模型预测SOC（R2 = 0.97, RMSE = 1.45 g/kg）、ROC（R2 = 0.94, RMSE = 0.55 g/kg）和MBC（R2 = 0.63, RMSE = 48.87 mg/kg）获得了最好的模拟结果；全谱FS-ANN预测DOC（R2 = 0.87, RMSE = 164.01 mg/kg）的精度最高。（2）采用CARS变量选择方法，SOC、DOC、ROC和MBC建模的关键特征波段从原始的1020个波段分别减少到70、53、52和37个；CARS变量优选不同程度地改善了模型的预测精度，其中CARS-SVMR对SOC、DOC和ROC三种组分均取得了最好的验证结果（R2 = 0.88–0.97），CARS-GPR对MBC（R2 = 0.51）的预测能力最好。（3）红壤性水稻土SOC主要以烷氧碳为主（30.0%–34.2%），其次为烷基碳（30.7%–32.6%）、芳香碳（19.2%–25.5%）和羰基碳（13.8%–14.0%），表层0–5 cm土壤比剖面下层55–60 cm土壤含有较多的烷基碳和烷氧碳。（4）土壤活性碳组分估算模型的预测精度在很大程度上取决于其组分含量与SOC含量或者土壤氧化铁含量之间的相关性程度。利用Vis-NIR高光谱成像技术能够实现原状土壤剖面不同碳组分的定量预测与精细制图，该方法可为剖面关键土壤属性空间分布的监测提供了新的理论和技术支持，从而更好地理解剖面SOC转化和累积的作用机制。