顾及土壤类型的土壤Zn含量高光谱遥感反演  

Soil Zn Content Inversion by Hyperspectral Remote Sensing Data and Considering Soil Types

作　　者：张霞 王为昊 孙伟超 丁松滔 王一博 ZHANG Xia;WANG Wei-hao;SUN Wei-chao;DING Song-tao;WANG Yi-bo(Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100101,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院空天信息创新研究院,北京100101 [2]中国科学院大学,北京100049

出　　处：《光谱学与光谱分析》2023年第7期2019-2026,共8页Spectroscopy and Spectral Analysis

基　　金：中国科学院战略性先导科技专项课题(XDA28080500);国家自然科学基金项目(42001282)资助。

摘　　要：目前针对土壤重金属的高光谱反演方法大多集中在单一的研究区域或未考虑土壤类型对反演结果的影响,而土壤类型和成土因素的不同会对土壤属性参数的高光谱反演模型的普适性产生一定程度影响。该研究提出一种顾及土壤类型的重金属高光谱遥感反演方法,根据研究区土壤类型,从土壤样本的实验室光谱中提取对重金属起主要吸附作用的土壤光谱活性物质的特征谱段,分别建立基于土壤光谱活性物质特征谱段的重金属含量估算模型。使用改进的遗传算法(IGA)对特征谱段进行波段优选,使用偏最小二乘回归算法(PLSR)建模,使用决定系数(R^(2))、相对偏差(RPD)和预测均方根误差(RMSEP)三个指标对不同的建模方法进行评价。以湖南省郴州市东河流域铅锌矿矿区的黄壤和红壤样本数据为例,采集38个黄壤样本和35个红壤样本,从土壤样本的实验室光谱中提取对Zn起主要吸附作用的土壤有机质和黏土矿物的特征谱段,均采用IGA+PLSR方法进行建模。结果表明:不考虑土壤类型即利用全部土壤样本进行建模时,与全谱段建模结果相比,基于土壤有机质和黏土矿物特征谱段的重金属Zn含量反演精度的R^(2)由0.624提升到0.755,RPD由1.668提升到2.069,RMSEP减少40.591;与不考虑土壤类型的建模相比,黄壤样本特征谱段的估算精度R^(2)由0.761提升到0.879,RPD由2.137提升到3.001,RMSEP减少74.737,红壤样本特征谱段的估算精度R^(2)由0.866提升到0.939,RPD由2.848提升到4.212,RMSEP减少89.358,黄壤和红壤样本的反演模型均达到了出色模型的标准。因此,土壤光谱活性物质特征谱段的提取以及土壤类型的考虑均有助于提高土壤Zn含量的反演精度,为应用高光谱遥感图像进行大范围土壤重金属污染监测奠定方法基础。At present,the research of hyperspectral inversion methods for heavy metals is mostly focused on single areas or not consider the influence of soil type on the inversion results.However,the differences of soil type and soil forming factors may have a certain degree in influence on the applicability of hyperspectral data-based inversion model of soil parameters.Hyperspectral remote sensing data proposed a method for inverting soil Zn metal content and considering soil types.It extracted the characteristic spectrum of soil spectrally active constituents with strong sorption and retention for the heavy metal from laboratory spectra of soil samples to enhance the inversion mechanism.For each soil type,the improved genetic algorithm(IGA)was employed on the characteristic spectrum to select the effective bands furtherly,and these bands were used to construct model by the partial least squares regression(PLSR).Finally,different modeling methods are evaluated using the coefficient of determination(R^(2)),relative deviation(RPD)and root mean square error of prediction(RMSEP).The proposed method was validated by 38 yellow soil samples and 35 red soil samples collected in the Dong River lead-zinc mining area in Chenzhou City,Hunan province,and then the soil Zn content inversion model of each of the two soil types was constructed by the characteristic spectrum of organic matter and clay minerals extracted from the laboratory spectra and finally both were modeled by the IGA+PLSR.The results showed,when modeling with all soil samples regardless of soil types,comparing with inversion using the entire spectral range,the R^(2) and RPD were improved from 0.624 and 1.668 to 0.755 and 2.069 and the RMSEP decreased by 40.591 by using the characteristic spectrum associated with soil organic matter and clay minerals.When considering soil types and modeling respectively,comparing with the inversion without considering soil types,for yellow soil,the R^(2) was improved from 0.761 to 0.879,RPD was improved from 2.137 to 3.001,and the RMSEP

关 键 词：重金属 土壤类型 高光谱遥感 土壤光谱活性物质 特征选择 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]