设为首页 加入收藏 登录旧版
空间关联随机森林模型结合Sentinel-2影像估算潍北地区裸土期土壤盐分
Estimation of Soil Salt Content in Bare Soil Period in Weibei Region Using Spatial Random Forest Model Combined with Sentinel-2 Imagery
投稿时间:2022-04-15  修订日期:2023-05-12
DOI:10.19316/j.issn.1002-6002.2023.05.25
中文关键词:  土壤盐分  Sentinel-2影像  空间关联随机森林模型  裸土期  潍北地区
英文关键词:soil salt content  Sentinel-2 image  spatial random forest model  bare soil period  Weibei region
基金项目:环境演变与自然灾害教育部重点实验室开放课题项目(2022-KF-14)
作者单位
彭远新 枣庄学院旅游与资源环境学院, 山东 枣庄 277160 
王泽强 枣庄学院旅游与资源环境学院, 山东 枣庄 277160 
周忠科 枣庄学院旅游与资源环境学院, 山东 枣庄 277160 
宋晓宁 枣庄学院旅游与资源环境学院, 山东 枣庄 277160 
徐夕博* 山东师范大学地理与环境学院, 山东 济南 250358 
通讯作者:徐夕博*  山东师范大学地理与环境学院, 山东 济南 250358  
摘要点击次数: 353
全文下载次数: 412
中文摘要:
      土壤盐分含量(SSC)是评价土地退化和肥力水平的重要指标,实现SSC状态和空间分异的快速准确监测对区域环境的优化管理极为关键。选取潍北平原为研究区,野外采集233处土壤样品并获取同时相Sentinel-2多光谱影像,进一步将特征光谱波段和构建的最优光谱指数作为输入自变量,测试得到的SSC实测值为因变量,最后将空间关联函数引入到随机森林中去建立基于空间关联随机森林算法的SSC遥感估算模型,完成区域尺度上的SSC反演估算与空间制图。结果表明:影像的B3、B8和B11是SSC的特征波段,通过波段比值变换能够增强卫星光谱信号对SSC的吸收响应,筛选得到的最优光谱指数分别为RI34(波段3和波段4的反射率比值)、RI711(波段7和波段11的反射率比值)、ND611(波段6和波段11的反射率归一化值)和D45(波段4和波段5的反射率差值);仅用特征波段或最优光谱指数来构建模型不能取得满意的SSC估算精度,空间关联随机森林模型的SSC估算精度要高于随机森林模型;在将上述特征波段和最优光谱指数共同输入空间关联随机森林遥感估算模型时,估算精度指标R2和相对分析误差RPD达到0.89和2.04,对比随机森林模型精度分别提升了28.99%和53.40%,能够较准确地完成区域SSC的估算制图;SSC在空间分布上中部略高于南部和北部,高值区主要受盐田分布影响。研究构建基于特征波段和最优光谱指数组合输入的空间关联随机森林模型,可为利用卫星遥感数据进行SSC的估算制图和区域内土壤环境的监测管理提供技术支持。
英文摘要:
      Soil salt content (SSC) is an important indicator for evaluating land degradation and fertility levels,it is critical to the regional environment management through realizing the rapid and accurate monitoring of the content status and spatial variation of SSC.The Weibei Plain was selected as the study area,233 soil samples were collected,and simultaneous Sentinel-2 multispectral images were acquired.Then the sensitive spectral bands and the constructed optimal spectral index were taken as input independent variables,and the measured value of the SSC obtained by the laboratory analysis as the dependent variable;Finally,the spatial correlation function was introduced into the random forest algorithm to establish the SSC remote sensing estimation model.SSC estimation and map were completed at a regional scale.The results showed B3,B8,and B11 were the sensitive bands of SSC,and B11 had the highest importance value of SSC.The spectral signal of SSC could be enhanced by band ratio transformation,and the optimal spectral indices obtained by screening were RI34 (reflectance ratio of band 3 and band 4),RI711 (reflectance ratio of band 7 and band 11),ND611 (reflectance normalization value of band 6 and band 11) and D45 (reflectance difference between band 4 and band 5).Only using sensitive bands or optimal spectral indices as inputs to build a model could not achieve satisfactory SSC estimations.The SSC estimation accuracy of the spatial random forest model was higher than that of the random forest model.When the above sensitive bands and the optimal spectral indices were jointly input into the spatial random forest remote sensing model,the estimation accuracy index R2 and RPD reached 0.89 and 2.04.Compared with the random forest model,the accuracy was increased by 28.99% and 53.40%,respectively.It could accurately complete the SSC estimates and maps at a regional scale.The SSC in the central part was slightly higher than that in the southern and northern parts,and the high-value areas were mainly affected by the saltern.The spatial random forest model based on the combination input of sensitive bands and optimal spectral indices could provide support for the SSC estimation and environmental monitoring at a regional scale.
查看全文  查看/发表评论  下载PDF阅读器