1、2023，45（2）DOI：10.13836/j.jjau.2023048Acta Agriculturae Universitatis Jiangxiensishttp:/江西农业大学学报 基于原位光谱的滨海滩涂土壤含水量预测模型宋敬茹1，满卫东1，2，3，4*，高均海5*，张永彬1，刘明月1，2，3，4，郝玉峰1，郑浩1，杨晓芜1（1.华北理工大学 矿业工程学院，河北 唐山 063210；.唐山市资源与环境遥感重点实验室，河北 唐山 063210；3.河北省矿区生态修复产业技术研究院，河北 唐山 063210；.河北省矿业开发与安全技术重点实验室，河北 唐山 063210；5.中煤科工生态
2、环境科技有限公司唐山分公司，河北 唐山 063012）摘要：【目的】滩涂土壤含水量对滨海区域生态保护和植物生长都至关重要，及时了解土壤含水量对生态恢复、土壤资源管理和土壤可持续利用具有重要意义。针对土壤含水量获取复杂的问题，探求更加快速精准地获取滨海滩涂土壤含水量的方法。【方法】沿沧州滨海区域采集了共计14个表层（020 cm）滩涂土壤样品，在实验室利用烘干法测得滩涂土壤含水量。在分析野外实测滨海滩涂土壤原位高光谱反射率（R）和土壤含水量特性的基础上，采用反射率倒数（1/R）、反射率倒数对数（log（1/R）、反射率一阶微分（R）和反射率去包络线（CR）等变换形式。充分挖掘光谱信息，并探求与滨
3、海滩涂土壤高相关性的可见近红外（VIS-NIR）光谱波段，构建基于多元逐步线性回归（MSR）和支持向量机回归（SVR）方法的滨海滩涂土壤含水量预测模型。结合适用于小样本的留一交叉验证（LOO-CV）法验证模型精度，并对比分析两种预测模型的性能以及模型的稳定性。【结果】研究发现：1）土壤光谱反射率与对应土壤含水量呈显著负相关，且二者在1 4001 600 nm和1 9002 400 nm内密切相关。2）除R外，同一种光谱变换形式下，采用SVR方法构建的滨海滩涂土壤含水量预测模型精度和稳定性明显高于MSR方法，对比得出基于R的滨海滩涂土壤含水量SVR预测模型精度最高，Adjusted-R2、RPD
4、和RMSE分别为0.81、2.08和2.56。【结论】在R变换形式下利用SVR方法建立的模型能够较准确地预测滨海滩涂土壤含水量，为滨海湿地土壤管理、植物生长和环境保护提供必要的数据支持，并为基于高光谱影像的区域尺度土壤含水量预测提供方法借鉴。关键词：高光谱；滨海滩涂；机器学习；土壤含水量；预测模型中图分类号：S152.7 文献标志码：A 开放科学（资源服务）标识码（OSID）：文章编号：1000-2286（2023）02-0508-09收稿日期：20220531 修回日期：20220930基金项目：国家自然科学基金青年科学基金项目（42101393，41901375）和河北省自然科学基金项目（
5、D2019209322，D2022209005）Project supported by the Youth Science Fund Project of National Natural Science Foundation of China（42101393，41901375）and Natural Science Foundation of Hebei Province（D2019209322，D2022209005）作者简介：宋敬茹，硕士生，orcid.org/0000-0003-3355-191X，；*通信作者：满卫东，讲师，主要从事生态遥感研究，orcid.org/0000-00
6、03-1960-1976，；高均海，研究员，主要从事矿山生态环境监测与矿山损害防治方面的研究，orcid.org/0000-0009-0000-4106-085，。宋敬茹，满卫东，高均海，等.基于原位光谱的滨海滩涂土壤含水量预测模型 J.江西农业大学学报，2023，45（2）：508-516.SONG J R,MAN W D,GAO J H,et al.Prediction model of soil water content in tidal flats based on in-situ spectrumJ.Acta agriculturae universitatis Jiangxien
7、sis,2023,45(2):508-516.第 2 期宋敬茹等:基于原位光谱的滨海滩涂土壤含水量预测模型Prediction Model of Soil Water Content in Tidal Flats Based on In-Situ SpectrumSONG Jingru1，MAN Weidong1,2,3,4*，GAO Junhai5，ZHANG Yongbin1，LIU Mingyue1,2,3,4，HAO Yufeng1，ZHENG Hao1，YANG Xiaowu1（1.College of Mining Engineering，North China Universit
8、y of Science and Technology，Tangshan，Hebei 063210，China；2.Tangshan Key Laboratory of Resources and Environmental Remote Sensing，Tangshan，Hebei 063210，China；3.Hebei Industrial Technology Institute of Mine Ecological Restoration，Tangshan，Hebei 063210，China；4.Hebei Key Laboratory of Mining Development
9、and Security Technology，Tangshan，Hebei 063210，China；5.Tangshan Branch，CCTEG Ecological Environment Technology Co.，Ltd.，Tangshan，Hebei 063012，China）Abstract：Objective The soil moisture content of the tidal flats is of great significance for the ecological protection，the plant growth of coastal area，e
10、cological restoration，soil resource management and sustainable soil use.In view of the complicated problem of soil moisture content acquisition，the method of obtaining soil moisture content of coastal tidal flats more rapidly and accurately is explored.Method A total of 14 surface（0-20 cm）tidal soil
11、 samples were collected along the coastal area of Cangzhou，and the soil moisture content of the tidal flats was measured by drying method in the laboratory.On the basis of analyzing the in-situ hyperspectral reflectance（R）and soil moisture content characteristics of the soil measured in the field，th
12、e spectral information and the visible-near-infrared（VIS-NIR）spectral band with high correlation with the coastal tidal flats soil were explored with the reflectance reciprocal（1/R），the reciprocal logarithm of reflectance（log（1/R），the first-order differential of reflectance（R）and the removal continu
13、um of reflectance（CR）.A prediction model of the soil moisture content of the coastal tidal flats soil based on multiple stepwise linear regression（MSR）and support vector machine regression（SVR）methods were constructed.The accuracy of the model was verified by combining the Leave-One-Out Cross-Valida
14、tion（LOO-CV）method for small samples，and the performance of the two predictive models and the stability of the models were compared and analyzed.ResultThe results show that：1）The spectral reflectance of soil was significantly and negatively correlated with the corresponding soil moisture content，and
15、 the two were closely related in the spectral ranges of 1 400-1 600 nm and 1 900-2 400 nm.2）In addition to R，the accuracy and stability of the prediction model of soil moisture content of coastal tidal flats constructed by SVR method was significantly higher than that of MSR method in the same spect
16、ral transformation form，and the accuracy and stability of SVR prediction model of soil moisture content based on R was the highest，with Adjusted-R2，RPD and RMSE of 0.81，2.08 and 2.56.ConclusionThe model established by SVR method under R transform can accurately predict the soil moisture content of coastal tidal flats，which provides data support for soil management，plant growth and environmental protection of coastal wetlands，and methods reference for predicting soil moisture content at regional
