黄河中游退耕还林地土壤有机碳含量的高光谱估测--以大宁县为例  

Estimation of Soil Organic Carbon in Returning Cropland to Forest in the Middle Reaches of the Yellow River Based on Hyperspectral Data-Take Daning County as an Example

作　　者：邓永鹏 朱洪芬[1] 丁皓希[1] 孙瑞鹏 毕如田[1] DENG Yongpeng;ZHU Hongfen;DING Haoxi;SUN Ruipeng;BI Rutian(College of Resource and Environment,Shanxi Agricultural University,Taigu 030801,China)

机构地区：[1]山西农业大学资源环境学院,山西太谷030801

出　　处：《山西农业科学》2022年第6期869-877,共9页Journal of Shanxi Agricultural Sciences

基　　金：山西省回国留学人员科研资助项目(2017-075)。

摘　　要：为了快速获取退耕还林地土壤有机碳含量数据,明确退耕还林工程的实施效果,以黄河中游大宁县退耕还林土壤为研究对象,获取土壤有机碳及光谱曲线数据;并选择原始光谱及其倒数的对数、倒数的对数一阶微分、一阶微分、去包络线5种光谱数据作为自变量,首先与土壤有机碳含量进行相关分析,选取特征波段,然后分别建立主成分回归、偏最小二乘回归和支持向量回归3种土壤有机碳高光谱估测模型。结果表明,土壤有机碳含量与光谱反射率呈负相关,即有机碳含量越高反射率越低,光谱曲线总体上呈现递增的趋势,在可见光范围内反射率增长速度较快,近红外范围内增长速度缓慢;不同光谱变换形式可以提高土壤有机碳含量与光谱反射率的相关性,其中倒数的对数一阶微分和去包络线光谱提升效果最好。分析不同光谱变换形式的建模精度发现,同一光谱数据在不同模型中建模精度存在显著差异,同时对比3种建模方法发现,支持向量回归方法精度较好,以倒数的对数一阶微分为自变量的支持向量回归模型精度最高,建模集和验证集的R2分别为0.780、0.707。高光谱技术可以准确、快速地进行土壤有机碳含量的估算。The purpose of this study is to explore how to quickly obtain data of the content of soil organic carbon in areas of returning cropland to forest in order to clarify the implementation effect of returning cropland to forest project.In this study,the areas with returning cropland to forest in Daning county in the middle reaches of the Yellow River were selected to obtain the data of the content of soil organic carbon and soil spectral curves.Five forms of spectral data were selected as independent variables,including the original spectrum,its reciprocal logarithm,reciprocal logarithmic′s first-order differential,first-order differential,and continuum removal.Firstly,correlation analysis between the content of soil organic carbon and soil spectrum was carried out.Then,three hyperspectral estimation models of soil organic carbon of principal component regression,partial least squares regression,and support vector regression were established by selecting characteristic bands.The results showed that the content of soil organic carbon was negatively correlated with spectral reflectance,as the content of soil organic carbon increased,the spectral reflectance decreased,the spectral curve showed an increasing trend in general,the reflectivity increased rapidly in the visible range and slowly in the near infrared range;the correlation between the content of soil organic carbon and spectral reflectance could be improved by the spectral transformation,the reciprocal logarithmic first-order differential and continuum removal had the best spectral enhancement effect.By analyzing the modeling accuracy of different spectral transformation forms,it was found that the modeling accuracy of the same spectral data was significantly different in different models.In comparison of three modeling methods,it was found that the support vector regression model had higher accuracy;the model of support vector regression with first-order differential of the reciprocal logarithmic as the independent variable had the highest accuracy,and R2 of

关 键 词：土壤有机碳 高光谱 支持向量回归 退耕还林地 黄河中游 

分 类 号：S153.62[农业科学—土壤学]