基于随机森林算法的贺兰山东麓洪积扇微地形分类研究  

Research on the classification of micro-terrain in the alluvial fan of Helan Mountains east foot based on Random Forest algorithm

作　　者：李红霞 石云[1] 沈爱红 马益婷 梁咏亮 李晓娟 董军 赵娜[1] 佘洁 王彤 张风红 郭瑞 LI Hongxia;SHI Yun;SHEN Aihong;MA Yiting;LIANG Yongliang;LI Xiaojuan;DONG Jun;ZHAO Na;SHE Jie;WANG Tong;ZHANG Fenghong;GUO Rui(College of Geographical Sciences and Planning,Ningxia University,Yinchuan 750021,China;College of Forestry and Prataculture,Ningxia University,Yinchuan 750021,China;Administration Bureau of Helan Mountain National Nature Reserve,Yinchuan 750021,China;Institute of Survey and Mapping,Yinchuan 750011,China;Yinxi Ecological Shelter Forest Management and Protection Center,Yinchuan 750002,China)

机构地区：[1]宁夏大学地理科学与规划学院,银川750021 [2]宁夏大学林业与草业学院,银川750021 [3]宁夏贺兰山国家级自然保护区管理局,银川750021 [4]银川市勘察测绘院,银川750011 [5]银川市银西生态防护林管护中心,银川750002

出　　处：《第四纪研究》2025年第1期178-190,共13页Quaternary Sciences

基　　金：宁夏回族自治区自然科学基金重点项目(批准号:2022AAC02020);中国工程院院地合作重大战略研究项目(批准号:2021NXZD8)共同资助。

摘　　要：微地形影响水热分布,形成不同的表面砾石、灌木、草本等组成分布特征,提取洪积扇微地形,构建洪积扇区微地形数字分类体系,可以为洪积扇地区土壤水热空间异质性、植物群落的空间分布、水源涵养能力以及土地利用分析提供依据。本研究以贺兰山东麓洪积扇为研究区,根据其地表形态、相对高差、砾石粒径大小以及植被组成,将洪积扇划分为冲积台地、高漫滩、冲沟、槽滩等4种微地形分类体系;以高精度DEM数据和无人机遥感影像为数据源,应用面向对象技术与数字地形分析结合的微地形识别与分类的方法,提取微地形的光谱、地形、纹理、几何等特征信息,通过影像分割、特征优选、随机森林(RF)分类算法,对贺兰山东麓洪积扇微地形进行识别分类,并验证分类精度。结果表明:1)融合地形特征的微地形面向对象分类的最优分割尺度为35;2)光谱特征和地形特征在贺兰山洪积扇微地形分类时重要性程度高,对微地形识别贡献度大;3)RF对微地形识别分类的效果最佳,总体精度为89.17%, Kappa系数为0.8480;4)微地形空间分布以高漫滩为主,广泛分布于洪积扇顶部,面积0.1224 km^(2),约占整个研究区总面积51.09%,冲积台地和槽滩的面积为0.0484 km^(2)、 0.0528 km^(2),分别占总面积的20.24%、 22.23%;冲沟分布最少,仅占总面积6.42%。总体上微地形的空间格局呈现地形类别之间交叉镶嵌分布特征。研究结果可以用于分析贺兰山洪积扇地区的微地形空间分布格局差异,为荒漠草原生态系统中地形复杂地区的洪积扇生态本底环境监测与水源涵养提供科学依据,对贺兰山东麓洪积扇生态系统的合理利用与科学管理具有重要意义。Micro-topography affects the distribution of water and heat,forming different surface gravel,shrubs,herbs,and other compositional distribution characteristics.Extracting the micro-topography of the alluvial fan and constructing a digital classification system for the micro-topography in the alluvial fan area can provide a basis for the spatial heterogeneity of soil water and heat,the spatial distribution of plant communities,water conservation capacity,and land use analysis in the alluvial fan region.This study takes the alluvial fan at the eastern foot of Helan Mountain as the research area and divides the alluvial fan into four types of micro-topography classification systems:alluvial terrace,high floodplain,gully,and channel beach,according to its surface morphology,relative height difference,gravel particle size,and vegetation composition.Using high-precision DEM data and unmanned aerial vehicle remote sensing images as data sources,the method of micro-topography identification and classification combines object-oriented technology with digital terrain analysis to extract spectral,topographical,textural,and geometric feature information of micro-topography.Through image segmentation,feature optimization,and the Random Forest(RF)classification algorithm,the micro-topography of the alluvial fan at the eastern foot of Helan Mountain is identified and classified,and the classification accuracy is verified.The results show that:(1)The optimal segmentation scale for object-oriented classification of micro-topography integrated with terrain features is 35;(2)Spectral and topographical features are highly important in the classification of micro-topography on the Helan Mountain alluvial fan,contributing significantly to the identification of micro-topography;(3)RF achieves the best effect on the identification and classification of micro-topography,with an overall accuracy of 89.17% and a Kappa coefficient of 0.8480;(4)The spatial distribution of micro-topography is dominated by high floodplains,which are widely dist

关 键 词：洪积扇 微地形 随机森林 数字地形分析 特征优选 

分 类 号：P23[天文地球—摄影测量与遥感] P931.2[天文地球—测绘科学与技术]