河南农业科学 ›› 2022, Vol. 51 ›› Issue (5): 162-170.DOI: 10.15933/j.cnki.1004-3268.2022.05.017

• 农业信息与工程·农产品加工 • 上一篇    下一篇

利用空间关联随机森林模型与遥感影像估算裸土期耕地土壤盐分含量的研究

徐夕博1,2,吕明荟3,4,王海会3,4,周忠科2,彭远新2,颜学文2   

  1. (1.北京师范大学地理科学学部,北京 100875;2.枣庄学院旅游与资源环境学院,山东 枣庄 277160;3.山东省地下水环境保护与修复工程技术研究中心,山东济南250014;4.山东省地质矿产勘查开发局八〇一水文地质工程地质大队,山东 济南 250014)
  • 收稿日期:2021-11-30 出版日期:2022-05-15 发布日期:2022-07-18
  • 作者简介:徐夕博(1994-),男,山东临沂人,在读博士研究生,研究方向:资源遥感。E-mail:xu_xibo@126.com
  • 基金资助:
    山东省教育厅公派出国留学基金项目(2019027)

Estimation of Soil Salt Content of Cultivated Land during Bare Soil Period Using Spatial Random Forest Model and Remote Sensing Images

XU Xibo1,2,LÜ Minghui3,4,WANG Haihui3,4,ZHOU Zhongke2,PENG Yuanxin2,YAN Xuewen2   

  1. (1.Faculty of Geographical Science,Beijing Normal University,Beijing 100875,China;2.College of Tourism,Resources and Environment,Zaozhuang University,Zaozhuang 277160,China;3.Shandong Engineering Research Center for Environmental Protection and Remediation on Groundwater,Ji’nan 250014,China;4.801 Institute of Hydrogeology and Engineering Geology,Shandong Provincial Bureau of Geology&Mineral Resources,Ji’nan 250014,China)
  • Received:2021-11-30 Published:2022-05-15 Online:2022-07-18

摘要: 为快速准确获取区域内土壤盐分含量(Soil salt content,SSC)信息及空间分布特征,选取莱州湾南岸滨海平原地区为研究区,系统采集裸土期土壤样品95个并获取同时相Sentinel-2多光谱影像,进一步利用变量重要度评估技术选取SSC的敏感波段作为输入自变量,测定得到的SSC值为因变量,分别建立基于随机森林和空间关联随机森林算法的遥感估算模型,完成区域尺度上的SSC反演制图。结果表明:Sentinel-2影像近红外范围内波段(B11、B12、B2和B8a)对SSC响应较为敏感,波段11的重要度值最高。空间关联随机森林模型的估算效果指标R2(决定系数)和RMSE(均方根误差)分别为0.86和0.38,相对比随机森林模型估算效果提升了16.22%和35.60%,模型的计算效率和稳健性也相应提高。SSC在区域上整体呈现较低含量水平(<2 g/kg),高值区分布在西北部和东部部分区域(>6 g/kg),主要受到海水入侵影响。此外,在微地貌和干旱蒸发的作用下形成的离散块状分布的中度土壤盐化区(2 g/kg≤SSC<4 g/kg)应引起重视。

关键词: 土壤, 盐分含量, 空间关联随机森林, 遥感估算, 裸土期

Abstract: This study aimed to investigate the concentration status and spatial distribution of soil salt content(SSC).The coastal plain area in the south of Laizhou Bay was selected as the study area,a total of 95 soil samples in bare soil period and Sentinel‑2 multispectral images were acquired simultaneously;the SSC responding bands were then selected using the variables importance assessment method,and used as the independent variables in the SSC estimation model,the measured SSC values were regarded as the dependent variables,and remote sensing estimation models based on random forest and spatial random forest were established.Results showed that the band 11,band 12,band 2,and band 8a from the Sentinel‑2 image were the sensitive bands for SSC,band 11 showed the significant correlation with SSC;The R2(coefficient of determination)and RMSE(root mean square error)of the spatial random forest model were 0.86 and 0.38,respectively,which were 16.22% and 35. 60% higher than those of the random forest model,and the computational efficiency and robustness of the model were improved accordingly.SSC showed a low level(<2 g/kg)at the regional scale,and high‑value areas(>6 g/kg)were mainly distributed in the northwestern and eastern parts of the area,mainly affected by seawater intrusion.In addition,moderately salt‑affected areas(2 g/kg≤SSC<4 g/kg)were characterized with discrete distribution due to the effects of micro‑geomorphology and drought evaporation,which should be given attention to.

Key words: Soil, Salt content, Spatial random forest, Remote sensing estimation, Bare soil period

中图分类号: