基于综合光谱指数的不同程度人类干扰下土壤有机质含量预测  

Prediction of soil organic matter content based on comprehensive spectral index at different levels of human disturbance

作　　者：郑曼迪 熊黑钢[2] 乔娟峰 刘靖朝 ZHENG Man-di;XIONG Hei-gang;QIAO Juan-feng;LIU Jing-chao(College of Resource and Environment Science,Xinjiang University/Key Laboratory of Oasis Ecdogy,Ministry of Education,Urumqi 830046,China;College of Art and Science,Beijing Union University,Beijing 100083,China)

机构地区：[1]新疆大学资源与环境科学学院/教育部绿洲生态重点实验室,新疆乌鲁木齐830046 [2]北京联合大学应用文理学院,北京100083

出　　处：《江苏农业学报》2018年第5期1048-1056,共9页Jiangsu Journal of Agricultural Sciences

基　　金：国家自然科学基金项目(41671198)

摘　　要：为寻求同一背景不同人类干扰程度下的土壤有机质含量的最佳预测模型,本研究以天山北麓的土壤为研究对象,运用Landsat8遥感影像以及实测光谱2种方式进行对比,结合不同的综合光谱指数,对无人干扰区、人为干扰区的影像反射率和实测光谱反射率进行光谱变换,分析反射率及其变换形式与有机质含量的相关性,以相关系数通过0. 01和0. 05显著性水平检验的波段作为自变量,运用多元线性回归方程分别建立了无人干扰区、人为干扰区土壤有机质含量高光谱预测模型,精度最高的为最优模型。结果表明:(1) Landsat8影像中B1～B5波段与有机质含量的相关系数通过了0. 01与0. 05显著性水平检验,作为自变量建立有机质含量预测模型。同时,为了能与影像反射率有个良好的对比,实测光谱反射率及其变换形式同样也选择5个相关系数最大的波段作为敏感波段用以建立模型。在影像与实测光谱中,土壤盐分指数结合植被指数与有机质含量相关性最好的分别是无人干扰区的SI_3、DVI和SI_3、RVI;人为干扰区的SI_2、RVI和SI_1、RVI。在结合光谱综合指数的模型中,无论是影像还是实测光谱,都是以反射率与植被指数、盐分指数相结合作为自变量建立的模型精度最好。对比2种预测方式,预测效果最好的是利用实测光谱与盐分指数、植被指数建立的无人干扰区一阶微分的多元线性回归模型以及人为干扰区的倒数之对数一阶微分的多元线性回归模型,R^2分别为0. 93和0. 89。To build the best prediction model of the soil organic matter contentunder the same background at different levels of human disturbance,using the soil of Fukang as the research object,Landsat8 remote sensing image and the measured spectra were used for comparison,and combined with different comprehensive spectral index to do spectral transformation between the image reflectance and the measured spectral reflectance.The correlation between reflectivity and its transformation and organic matter content was analyzed,the band with correlation coefficient through the significant test at 0.01 and 0.05 level was used as independent variable,the multivariate linear regression equations were used to establish high spectral prediction model of soil organic matter content.The results showed that the correlation coefficients of B1-B5 bands and organic matter content in Landsat8 images were tested at 0.01 and 0.05 significance level,and the bands were used as the independent variable to establish organic matter content prediction model.In order to form a good contrast with the results of image reflectivity,the five bands with the largest correlation coefficent were also selected as sensitive bands to establish prediction models of the measured spectral reflectance and its transformation form.The best correlation between soil salinity index combined with vegetation index and organic matter was SI 3,DVI and SI 3,RVI in the undisturbed zone,SI 2,RVI,and SI 1,RVI in human interference area.The model established by using reflectivity combined with vegetation index and salinity index as the independent variable had the best precision.The first order differential multivariate linear regression model established by measured spectrum and salinity index,vegetation index in unmanned interference area and first order differential multivariate linear regression model of the reciprocal in artificial interference region had best prediction effect,values of R 2 were 0.93 and 0.89,respectively.

关 键 词：干旱区 遥感 高光谱 土壤有机质 估算模型 

分 类 号：F301.24[经济管理—产业经济]