基于高光谱数据的盐荒地和耕地土壤盐分遥感反演优化  

Optimizing the inversion of soil salt in salinized wasteland using hyperspectral data from remote sensing

作　　者：孙亚楠 李仙岳[1] 史海滨[1] 马红雨 王维刚 崔佳琪 陈辰 Sun Yanan;Li Xianyue;Shi Haibin;Ma Hongyu;Wang Weigang;Cui Jiaqi;Chen Chen(College of Water Conservancy and Civil Engineering,Inner Mongolia Agricultural University,Hohhot 010018,China;Haolai Huresumu People's Government of Hexigten Banner,Chifeng 025374,China;Inner Mongolia Autonomous Region Water Resources Department Comprehensive Security Center,Hohhot 010010,China)

机构地区：[1]内蒙古农业大学水利与土木建筑工程学院,呼和浩特010018 [2]内蒙古赤峰市克什克腾旗浩来呼热苏木人民政府,赤峰025374 [3]内蒙古自治区水利厅综合保障中心,呼和浩特010010

出　　处：《农业工程学报》2022年第23期101-111,共11页Transactions of the Chinese Society of Agricultural Engineering

基　　金：十四五重点研发计划项目(2021YFC3201202);内蒙古科技计划项目(2022YFHH0039、2021CG0022)。

摘　　要：盐荒地作为研究区的“临时盐库”,其土壤盐分远高于研究区平均水平,因此探究不同土地利用类型土壤盐分的光谱响应差异以及对盐分遥感模型的影响,是实现不同土地类型土壤盐分反演值更加接近真实值的重要途径。该研究以河套灌区永济灌域为例,针对耕地和盐荒地土壤分别进行原位高光谱测定(FieldSpec 4 Hi-Res,ASD),对光谱数据进行多种光谱变换(基础数学变换、导数变换及光谱指数)后,分别基于特征波长和特征光谱指数构建单一土地类型盐分反演模型(耕地(Agricultural Land,AL)、盐荒地(Salinized Wasteland,SW))和整体盐分反演模型(耕地+盐荒地(Agricultural Land+Salinized Wasteland,AL+SW)),对比分析2种建模方式下的模型精度,提出区域土壤盐分遥感反演的最佳建模方式。结果表明:AL、SW和AL+SW中土壤样本数据的平均含盐量分别为5.09、13.42和7.09 g/kg,且在各等级盐分区间内,SW的光谱反射率均大于AL,其中轻度盐化土、中度盐化土和重度盐化土的光谱反射率平均差值分别为0.040、0.020和0.034;光谱变换和光谱指数均能有效改善不同土地类型中土壤盐分与光谱的相关性。相比基础变换(倒数、对数、根式等),导数变换不仅增大了敏感波长的范围,还使得特定波长处相关系数得到显著提升。不同土地类型中基于特征光谱指数的模型精度均高于基于特征波长的模型;单一土地类型盐渍化反演模型明显提高了区域土壤盐分的反演精度,单一土地类型盐渍化反演模型中(AL、SW模型)各变换下光谱指数模型平均决定系数相比整体模型(AL+SW模型)由0.50提高到了0.61,其中基础变换、一阶导数和二阶导数模型平均R2相比整体模型分别提高了0.06、0.11和0.17,同时,基于最优光谱指数的单一土地类型盐渍化反演模型平均R2相比整体模型由0.74提高到了0.92。因此,当区域中存在盐分相差较大的多种土地利用�Salinized wasteland can be served as the temporary salt reservoir with the much higher salt content beyond the average level.Therefore,the high-precision inversion of soil salinity can be realized to explore the difference in the spectral response of soil salinity in different land use types and its influence on the remote sensing model.In this study,a typical salinization region(Yongji of Hetao irrigation district in China)was chosen as the study region.The salinized wasteland was relatively scattered and mostly concentrated around the agricultural land.The salt content in the salinized wasteland was much higher than that in the agricultural land.Firstly,in-situ hyperspectral measurement(FieldSpec 4 Hi-Res,and ASD)was carried out for the agricultural land and salinized wasteland in April from 2018 to 2020.Secondly,the spectral data was subjected to the various spectral transformations,including the fundamental transformation(original,reciprocal,logarithm,and radical transformation),derivative transformation(the first and the second derivative),and spectral index(normalized differential soil index,difference soil index,and simple ratio soil indices),respectively.Thirdly,the multiple stepwise regressions were used to acquire the characteristic bands and spectral indices.Lastly,the single land type salt inversion model(Agricultural Land(AL),Salinized Wasteland(SW)),and the overall salt inversion model(Agricultural Land+Salinized Wasteland(AL+SW))were constructed using the characteristic wavelength and characteristic spectral index,respectively.The model accuracy under different modeling was evaluated using the coefficient of determination(R2),and Root Mean Square Error(RMSE).As such,the best modeling was proposed for the regional soil salinization.The results showed that the average content of soil salinity in the samples of AL,SW,and AL+SW model was 5.09,13.42,and 7.09 g/kg,respectively.Specifically,the SW spectral reflectance was greater than that of the AL in each wavelength range of different grades of salt zon

关 键 词：盐分 遥感 土壤 河套灌区 反演 光谱变换 盐荒地 

分 类 号：S127[农业科学—农业基础科学]