摘要
准确高效获取土壤水盐信息是盐碱地改良和可持续利用的前提。本研究以地面野外高光谱反射率和实测土壤水盐含量为数据源,利用分数阶微分(FOD)技术对原始光谱反射率进行步长为0.25的处理,从光谱数据与土壤水盐信息相关性层面筛选FOD阶数,构建二维光谱指数,采用支持向量机回归(SVR)和地理加权回归(GWR)建立土壤水盐含量反演模型并进行验证。结果表明:FOD技术可以在一定程度上减弱高光谱噪声并挖掘潜在光谱信息,提高高光谱反射率与土壤含水量(SMC)、pH值和含盐量的相关性,相关系数最高分别提升0.98、1.35和0.33。与一维光谱相比,FOD结合二维光谱指数筛选的特征波段组合对SMC、pH值和含盐量的响应更敏感,分别以1.5、1.0和0.75阶为最优,其中,SMC最大相关系数绝对值的最佳组合波段为570、1000、1010、1020、1330和2140 nm;pH值为550、1000、1380和2180 nm;含盐量为600、990、1600和1710 nm。相较于原始光谱反射率,SMC、pH值和含盐量最优阶次估算模型验证决定系数(R_(p)^(2))最高分别提升1.87、0.94和0.56。所建模型中GWR精度整体优于SVR,其中GWR最优阶次估算模型R_(p)^(2)分别为0.866、0.904和0.647,相对分析误差为3.54、4.25和1.86。研究区土壤含水量和含盐量总体呈西部低、东部高的空间分布特征,西北部土壤碱化问题较为严重,东北部较轻。研究结果可为引黄灌区土壤水盐高光谱反演提供科学依据,为盐碱地精准农业实施和管理提供新的策略。
Accurate and efficient acquisition of soil water and salt information is a prerequisite for the improvement and sustainable utilization of saline lands.With the ground field hyperspectral reflectance and the measured soil water-salt content as data sources,we used the fractional order differentiation(FOD)technique to process hyper-spectral data(with a step length of 0.25).The optimal FOD order was explored at the correlation level of spectral data and soil water-salt information.We constructed two-dimensional spectral index,support vector machine regres-sion(SVR)and geographically weighted regression(GWR).The inverse model of soil water-salt content was finally evaluated.The results showed that FOD technique could reduce the hyperspectral noise and explore the potential spectral information to a certain extent,improve the correlation between spectrum and characteristics,with the highest correlation coefficients of 0.98,1.35 and 0.33.The combination of characteristic bands screened by FOD and two-dimensional spectral index were more sensitive to characteristics than one-dimensional bands,with the opti-mal responses of order 1.5,1.0 and 0.75.The optimal combinations of bands for maximum absolute correction coef-ficient of SMC were 570,1000,1010,1020,1330 and 2140 nm,p H were 550,1000,1380 and 2180 nm,and salt content were 600,990,1600 and 1710 nm,respectively.Compared with the original spectral reflectance,the validation coefficients of determination(R_(p)^(2))of the optimal order estimation models for SMC,pH,and salinity were improved by 1.87,0.94 and 0.56,respectively.The overall GWR accuracy in the proposed model was better than SVR,where the GWR optimal order estimation models R_(p)^(2)were 0.866,0.904 and 0.647,and the relative per-centage difference were 3.54,4.25 and 1.86,respectively.The overall spatial distribution of soil water and salt con-tent in the study area was characterized by low in the west and high in the east,with more serious soil alkalinization problems in the northwest and less severe in the
作者
王怡婧
陈睿华
张俊华
丁启东
李小林
WANG Yijing;CHEN Ruihua;ZHANG Junhua;DING Qidong;LI Xiaolin(College of Geography and Planning,Ningxia University,Yinchuan 750021,China;Breeding Base for Sate Key Laboratory of Land Degrada-tion and Ecological Restoration in Northwest China/Ministry of Education Key Laboratory for Restoration and Recon-struction of Degraded Ecosystems in Northwest China,School of Ecology Environment,Ningxia University,Yinchuan 750021,China)
出处
《应用生态学报》
CAS
CSCD
北大核心
2023年第5期1384-1394,共11页
Chinese Journal of Applied Ecology
基金
国家重点研发计划项目(2021YFD1900602)
国家自然科学基金项目(42067003)
清华大学-宁夏银川水联网数字治水联合研究院联合开放基金项目(SKLHSE-2022-IOW11)资助。
关键词
野外高光谱
分数阶微分
二维光谱指数
地理加权回归
支持向量机回归
field hyperspectral
fractional order differentiation
two-dimensional spectral index
geographically weighted regression
support vector machine regression.
