丘陵区耕地土壤有机质含量高光谱估测研究  

Estimation of Soil Organic Matter Content of Cultivated Land in Hilly Regions with Hyperspectral

作　　者：于欢 刘健 刘亚秋[1] 顾子平 王瑷玲[1] YU Huan;LIU Jian;LIU Ya-qiu;GU Zi-ping;WANG Ai-ling(College of Resources and Environment/Shandong Agricultural University,Tai’an 271018,China)

机构地区：[1]山东农业大学资源与环境学院,山东泰安271018

出　　处：《山东农业大学学报（自然科学版）》2021年第4期648-653,共6页Journal of Shandong Agricultural University：Natural Science Edition

基　　金：山东省自然科学基金(ZR2019MD014);山东省重大科技创新工程项目(2018CXGC0209)。

摘　　要：本文以钢城区2个丘陵村耕地土壤为研究对象,通过野外采样、自然风干、化验分析、高光谱测定及数据处理等,确定最佳高光谱变换方式并筛选显著性波段,建立随机森林(RF)、支持向量机(SVM)、偏最小二乘回归(PLSR)和多元逐步回归(SMLR)4种估测模型,对比分析确定最佳估测模型。结果表明:高光谱变换处理可以扩大光谱曲线特征,提高与机质含量的相关性;一阶微分R’为最佳高光谱变换方式,筛选出706、1002、1359、1415、1886、1914和2221 nm 7个波段作为估测土壤有机质含量的显著性波段;建立的4种估测模型中,RF模型精度最高,其训练样本集R^(2)和RPD分别达到0.93、3.13,验证样本集R^(2)和RPD为0.73、1.87。因此,研究构建的R’-RF土壤有机质含量高光谱估测模型可为该丘陵区有机质含量的快速监测提供参考。The research selected the cultivated soil of two hilly villages in Gangcheng District as the research object,and through field sampling,natural air drying,laboratory analysis,hyperspectral measurement and data processing,etc,the best hyperspectral transformation method and the significant bands were selected.Random forest(RF),support vector machine(SVM),partial least squares regression(PLSR)and stepwise multiple linear regression(SMLR)estimation models were built,and through the ways of comparative analysis and accuracy evaluation,the best estimation model was selected.The results showed that by the process of hyperspectral conversion processing,the characteristics of the spectral curve can be expanded and the correlation with the soil organic content can be improved;the first-order differential is the best hyperspectral transformation method,and the seven bands of 706,1002,1359,1415,1886,1914 and 2221 nm were selected as the significant bands to estimate soil organic matter content;among the four estimation models,the RF model has the highest accuracy,the training sample set R^(2) and RPD reach respectively 0.93 and 3.13,and the verification sample set R^(2) and RPD are respectively 0.73 and 1.87.Therefore,the construction of the R’-RF soil organic matter content hyperspectral estimation model in this research can provide a reference for the rapid monitoring of the organic matter content in this hilly region.

关 键 词：土壤有机质 高光谱 模型估测 丘陵 

分 类 号：S15[农业科学—土壤学]