基于多光谱影像反演土壤盐分的建模方法研究  

Study on Model Method of Inversion of Soil Salt Based on Multispectral Image

作　　者：王明宽 莫宏伟[1] 陈红艳[2] 

机构地区：[1]湖南科技大学,湖南湘潭411201 [2]山东农业大学,山东泰安271000

出　　处：《土壤通报》2016年第5期1036-1041,共6页Chinese Journal of Soil Science

基　　金：湖南省教育厅科学研究项目(14C0467);地理空间信息技术国家地方联合工程实验室开放基金资助项目(2014GISNELJJ004)资助

摘　　要：土壤盐分的定量遥感反演,为快速、准确、全面地监测盐渍化状况提供了可能。本文以黄河三角洲地区垦利县为例,实地调查采集土壤样本,并获取同时相Landsat 8影像,建立土壤盐分遥感反演的BP神经网络、偏最小二乘回归、主成分分析、多元线性回归多种模型,进而进行精度对比分析,评价、优选最佳建模方法,最后,基于最佳模型进行研究区土壤盐分的空间分布反演分析。结果显示:遥感影像的反射率与土壤盐分含量并不是单纯的线性关系,构建的盐分估测模型BP神经网络预测决定系数为0.8467,均方根误差为0.071,明显高于传统线性统计模型,能较好地模拟土壤盐分与光谱数据的关系。该研究既能为盐渍土的治理、利用提供数据支持,又能推动盐渍化区域遥感研究的定量发展。Quantitative remote sensing inversion of soil salt makes it possible for monitoring salinization status rapidly, accurately and comprehensively. This paper took KenliCountry in Yellow River delta as a case, surveyedand collectedsoil samples fromthe field,and gotthe Landsat 8 image. The remote sensing inversion modelsof soil salinitywas built using BP neural network, partial least-squares regression, principal component analysis and multiple linear regression methods, the accuracy was compared and analyzed, and then the best model method was evaluated and selected. Finally, the spatial distribution of soil sahwas back analyzedbased on the best model in the study area. Results showed that the reflectivity of remote sensing image was not a simple linear relationship with soil salt content, R2and RMSEof salt estimation model based on BP neural network, was, respectively, 0.8467 and 0.071, which was higher significantly than that of the traditional statistical model.So the relationship between soil salt andthe spectral data couldbe well simulated by BP neural network.The research not only provided the datasupports for the management and the use of saline soil but also promoted the development of quantitative remote sensingin salinization area.

关 键 词：盐碱化 遥感反演 BP神经网络 偏最小二乘回归 主成分分析 多元线性回归 

分 类 号：S156.4[农业科学—土壤学]