利用U-Net的格陵兰冰盖冰面湖提取和面积变化分析  

Extraction and Area Change Analysis of Supraglacial Lakes in Greenland Ice Sheet Using U-Net Model

作　　者：彭博洋 周春霞[1,2] 朱冬雨 李明慈 肖倩雨 江萌 PENG Boyang;ZHOU Chunxia;ZHU Dongyu;LI Mingci;XIAO Qianyu;JIANG Meng(Chinese Antarctic Center of Surveying and Mapping,Wuhan University,Wuhan 430079,China;Key Laboratory of Polar Surveying and Mapping,Ministry of Natural Resources,Wuhan 430079,China)

机构地区：[1]武汉大学中国南极测绘研究中心,湖北武汉430079 [2]自然资源部极地测绘科学重点实验室,湖北武汉430079

出　　处：《武汉大学学报（信息科学版）》2024年第9期1621-1629,共9页Geomatics and Information Science of Wuhan University

基　　金：国家重点研发计划(2018YFC1406102);国家自然科学基金(42171133,41941010);湖北省自然科学基金(2019CFA057)。

摘　　要：冰面湖作为冰盖质量损失、海平面上升、气候变化等全球变化问题的指示器,对研究全球气候变化背景下冰盖的稳定性有着重要的作用。利用深度学习中的语义分割模型U-Net实现了不同情形下Landsat卫星影像中冰面湖的提取,并与多种方法进行了比较,发现U-Net冰面湖提取方法在性能和效率上表现最佳。提取2000—2020年北极格陵兰岛Sermeq Avannarleq冰川附近的冰面湖,并分析冰面湖面积的季节变化和年际变化。结果表明,该区域的冰面湖面积在每年5月—9月呈现先增后减的变化趋势,5月中旬开始增加,7月中下旬达到峰值,9月基本消亡;该区域的冰面湖面积在2000—2020年呈现增加的趋势,表明该区域冰盖融化量逐渐增加。Objectives:As a sensitive indicator of global changes such as ice sheet mass loss,sea level rise and climate change,supraglacial lakes play an important role in the study of ice sheet stability.Methods:U-Net is a semantic segmentation model,which extracts supraglacial lakes from Landsat satellite images in different situations.Results:The extraction of supraglacial lakes by U-Net possesses the optimal performance and efficiency.And the supraglacial lakes near the Sermeq Avannarleq Glacier in Greenland ice sheet are extracted from 2000 to 2020,and the seasonal and interannual variations of the area are analyzed.Conclusions:The area of supraglacial lakes starts to increase in mid-May,reaches the peak in the midlate-July,and almost disappears in September in a year.For the long time series change,the area of supraglacial lakes in this region shows an increasing trend from 2000 to 2020,indicating that the amount of ice sheet melting is gradually increasing.

关 键 词：冰面湖 格陵兰冰盖 深度学习 Landsat影像 

分 类 号：P237[天文地球—摄影测量与遥感]