东北典型黑土区表层土壤有机质含量高光谱反演研究  

Hyperspectral Retrieval of Surface Soil Organic Matter Content in a Typical Black Soil Region of Northeast China

作　　者：聂哲 李秀芬[1,3] 吕家欣 郑晓[2,3,4] NIE Zhe;LI Xiu-fen;LV Jia-xin;ZHENG Xiao(She nyang A gric ultural University,Shenyang 110866,China;Key Lab oratory of Forest Ecology and Management,Institute of Applied Ecology,Chinese Academy of Sciences,Shenyang 110016,China;Qingyuan Forest CERN,Chinese Academy of Sciences,Shenyang 110016,China;Daqinggou Ecological Station,Institute of Applied Ecology,Chinese Academy of Sciences,Shenyang 110016,China)

机构地区：[1]沈阳农业大学,辽宁沈阳110866 [2]中国科学院森林生态与管理重点实验室(沈阳应用生态研究所),辽宁沈阳110016 [3]中国科学院清原森林生态系统观测研究站,辽宁沈阳110016 [4]中国科学院沈阳应用生态研究所大青沟沙地生态实验站,辽宁沈阳110016

出　　处：《土壤通报》2019年第6期1285-1293,共9页Chinese Journal of Soil Science

基　　金：中国科学院前沿科学重点研究项目（QYZDJ-SSW-DQC027）;国家自然科学基金项目（31770758）资助

摘　　要：选择东北典型黑土区--德惠市、扶余市和双城市为研究区,利用便携式地物光谱仪获取土壤光谱数据,基于原始光谱值及一阶微分、倒数的对数、连续统去除变换,分别建立了黑土有机质含量的多元线性逐步回归模型、偏最小二乘回归模型和BP神经网络模型。结果表明:高光谱与土壤有机质含量在多个波段相关性较好,其中有机质与反射率一阶微分处理的相关性最好,在光谱584 nm处其相关性最强(r=-0.60,n=81)。光谱一阶微分处理数据在三种建模方法中的预测及验证精度均高于原始光谱值、倒数的对数和连续统去除变换,因此一阶微分为最佳光谱变换形式。偏最小二乘回归分析的预测效果整体优于多元线性逐步回归分析和BP神经网络分析,光谱一阶微分处理的偏最小二乘回归模型呈现出最佳预测效果,决定系数为0.71、均方根误差为2.29 g kg^-1(n=53)。Soil hyperspectral data were obtained by portable spectrometer in Dehui,Fuyu and Shuangcheng in typical Black Soil areas of Northeast China.Based on the original spectral values and the first-order differential,the logarithm of the reciprocal,and the continuum removal transformation,the models of stepwise multiple linear regression,partial least squares regression and BP neural network were established,respectively.The hyperspectral data and soil organic matter content showed a good correlation in multiple bands,among which soil organic matter was highly relevant with first-order differential treatment of reflectance,and the correlation was the strongest at 584 nm(r=-0.60,n=81).The prediction and verification accuracy of the first-order differential processing with the three modeling methods exceeded the original spectral values,the logarithm of the reciprocal and the continuum removal transformation,so the first-order differential was the optimum form of spectral transformation.The prediction effect by using the partial least squares regression analysis was better than that by using the stepwise multiple linear regression analysis and BP neural network analysis.The partial least squares regression model of spectral first-order differential treatment showed the best prediction effect,with coefficient of determination of 0.71 and the root mean square error of 2.29 g kg^-1(n=53).

关 键 词：土壤有机质 高光谱 模型估测 

分 类 号：S15[农业科学—土壤学]