基于GEE云平台的小江流域泥石流迹地空间分布制图  

Spatial Distribution Mapping of Debris Flow Site in Xiaojiang River Basin Based on the GEE Platform

作　　者：宗慧琳 袁希平 甘淑[1,2] 杨明龙 吕杰[1] 张晓伦[1] ZONG Hui-lin;YUAN Xi-ping;GAN Shu;YANG Ming-long;L Jie;ZHANG Xiao-lun(Faculty of Land Resource Engineering,Kunming University of Science and Technology,Kunming 650093,China;Research Center of Applied Engineering of Spatial Information Surveying and Mapping Technology in Plateau and Mountainous in Yunnan Province,Kunming 650093,China;Key Laboratory of Mountain Land Cloud Data Processing and Application for Universities in Yunnan Province,West Yunnan University of Applied Sciences,Dali 671006,China)

机构地区：[1]昆明理工大学国土资源工程学院,云南昆明650093 [2]云南省高校高原山地空间信息测绘技术应用工程研究中心,云南昆明650093 [3]滇西应用技术大学云南省高校山地实景点云数据处理及应用重点实验室,云南大理671006

出　　处：《光谱学与光谱分析》2025年第4期1045-1060,共16页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(62266026,41861054)资助。

摘　　要：快速、准确、详尽调研泥石流孕灾区域的分布信息能够帮助了解、深刻认识泥石流分布范围、分布规律及成因,并进一步根据具体情况找到科学的监测、预测、预防和治理的技术手段,从而减少泥石流灾害带来的问题与损失。为寻求高效、高精度的泥石流空间分布提取方法,以云南省小江流域作为研究区,利用谷歌地球引擎(Google Earth Engine,GEE)平台和随机森林算法,有效地提取了泥石流迹地的空间分布。首先利用2022年Sentinel-2影像及地形数据构建4类特征变量(光谱特征、指数特征、地形特征、纹理特征)作为特征集合,接着将随机森林特征变量重要性评分和J-M距离结合进行特征优选研究与分析,探讨了各个特征变量对泥石流迹地提取的重要性;最后设置不同特征组合形成6种不同的提取方案,对比分析6种试验方案提取泥石流迹地的精度,确定最优方案以提高识别精度。研究表明:(1)无论是否进行特征优选,加入地形特征变量的泥石流迹地提取精度均优于仅使用光学影像数据的精度,可见地形数据有利于泥石流迹地信息提取;(2)不同类型的特征变量对分类精度的影响不同,特征重要性评分由高到低的特征类型为地形特征、指数特征、纹理特征、光谱特征;(3)基于Sentinel-2光学影像和地形数据的多源数据构建多维特征变量并进行特征优选的试验方案6,提取到的2022年云南省小江流域泥石流迹地空间分布图最优,总体精度为94.95%,Kappa系数为0.94,泥石流迹地的制图精度为91.01%,用户精度为95.29%,该方案不仅提高了分类精度还有效降低了数据冗余。利用Google Earth Engine平台,光学遥感影像和地形数据相结合的多源数据以及随机森林算法,能够快速、准确、高效地制作较大范围地物覆盖复杂地区的泥石流迹地空间分布图,具有较大的应用潜力。Rapid,accurate and exhaustive research on the distribution of mudslide-hostile areas is of great significance,enabling us to understand and have a deep understanding of the scope of distribution of mudslides,the distribution pattern,the causes of the mudslides,and to further find scientific monitoring,prediction,prevention and management of the technical means by the specific situation,to reduce the problems and losses brought about by the mudslide disaster.To seek an efficient and high-precision method for extracting the spatial distribution of mudslides,this study chooses the Xiaojiang River Basin in Yunnan Province as the study area,employed the random forest algorithm based on the Google Earth Engine(GEE)platform to extract the spatial distribution of debris flow traces efficiently.Firstly,Four types of feature variables(spectral features,index features,topographic features,and texture features)were constructed using the 2022 Sentinel-2 image and topographic data,then the random forest feature variable importance score and the J-M distance were combined for the feature preference research and analysis,explored the importance of each feature variable on the extraction of mudslide traces,and finally,set up various feature combinations to create six schemes,compared and analyzed the accuracy of the debris flow traces extracted by the six experimental schemes,and found the best scheme to increase the recognition accuracy.The study shows that:(1)regardless of feature optimization,the accuracy of debris flow trace identification with the addition of terrain feature variables is higher than that with merely optical image data,indicating the utility of using terrain data for debris flow trace information extraction;(2)classification accuracy is affected differently by different feature variable kinds;topographic,index,texture,and spectral features are the feature types with the highest to lowest feature importance scores;(3)the experimental scheme 6 is the best results of the spatial distribution map of debris flow t

关 键 词：泥石流区提取 特征优选 J-M距离 Google Earth Engine Sentinel-2数据 随机森林 特征变量重要性 

分 类 号：P237[天文地球—摄影测量与遥感]