基于决策树的天山冰湖提取方法研究  

Research on the extraction method of Tianshan glacier lake based on decision tree

作　　者：李梦帆 郑江华[1,2] 钱安良 李家辉 阿迪力江·帕尔合提 王哲 马丽莎[1,2] 王南 LI Mengfan;ZHENG Jianghua;QIAN Anliang;LI Jiahui;Adiljan PARHAT;WANG Zhe;MA Lisha;WANG Nan(College of Geography and Remote Sensing Sciences,Xinjiang University,Urumqi 830017,Xinjiang,China;Xinjiang Key Laboratory of Oasis Ecology,Xinjiang University,Urumqi 830017,Xinjiang,China)

机构地区：[1]新疆大学地理与遥感科学学院,新疆乌鲁木齐830017 [2]新疆绿洲生态自治区重点实验室,新疆乌鲁木齐830017

出　　处：《干旱区研究》2024年第10期1699-1707,共9页Arid Zone Research

基　　金：国家级大学生创新研究项目(202210755003);第三次新疆综合科学考察项目(2021xjkk1001)资助。

摘　　要：天山位于亚欧大陆中部,是现代冰川的主要分布区之一,该地区冰川融水形成了数量多且分布广泛的冰湖。冰湖是气候变化的重要指示器,也是中国西北干旱与半干旱地区重要的地表水及地下水供给来源。由于地形因素和地物光谱特征的影响,使用单一的水体指数进行遥感影像的冰湖提取时,难以较好地区分出冰湖、山体阴影和积雪。本研究以天山地区为研究区,基于Google Earth Engine云平台,以Landsat 8遥感影像为数据源,根据冰湖的空间位置(缓冲区范围)、地形特征(坡度、高程)以及光谱特征,构建了冰湖决策树提取方法,并与NDWI(归一化水体指数)、MNDWI(改进的归一化水体指数)阈值法进行了精度比较。结果表明:决策树法能够有效减小山体阴影和积雪影响,更有效地提取冰湖信息,提取结果总体精度为89.14%,Kappa系数为0.783,F1分数为87.85%。结合了空间位置、地形特征和光谱特征的决策树方法为冰湖的动态监测与研究分析提供了一种较为高效的提取方法。The Tianshan region,located in the middle of the Eurasian continent,is a major distribution area of modern glaciers,and its glacial meltwater has formed a large number of and widespread glacial lakes.Glacial lakes are vital indicators of climate change and an important source of surface and groundwater supply in arid and semi-arid regions of Northwest China.The impact of topographic factors and spectral characteristics of ground objects makes it difficult to distinguish between glacial lakes,mountain shadow,and snow cover when extracting glacial lakes from remote sensing images using a single water index.In this study,based on the Google Earth Engine platform and Landsat 8 remote sensing images as the data source,a decision tree extraction method of the glacial lake was constructed according to the topographic characteristics(slope,elevation,and buffer analysis)and spectral characteristics of the glacial lake.This method was compared with the NDWI and MNDWI threshold methods.Experimental results demonstrated that the decision tree method can effectively reduce the impact of mountain shadow and snow cover and accurately extract glacial lake information.The overall accuracy of the extraction results was 89.14%,the Kappa coefficient was 0.783,and the F1 score was 87.85%.The decision tree method,which combines spatial topographic features and spectral features,is a relatively efficient extraction method for dynamic monitoring and research analysis of glacial lakes.

关 键 词：冰湖提取 决策树 Google Earth Engine 天山 

分 类 号：P343.6[天文地球—水文科学] P237[天文地球—地球物理学]