基于Sentinel-2 MSI与Sentinel-1 SAR相结合的黄土高原西部撂荒地提取——以青海民和县为例  

Sentinel-2 MSI and Sentinel-1 SAR based information extraction of abandoned land in the western Loess Plateau:A case study of Minhe County in Qinghai

作　　者：张昊 高小红[1,2,3,4] 史飞飞 李润祥 ZHANG Hao;GAO Xiaohong;SHI Feifei;LI Runxiang(School of Geographical Sciences,Qinghai Normal University,Xining 810008,China;MOE Key Laboratory of Tibetan Plateau Land Surface Processes and Ecological,Xining 810008,China;Qinghai Province Key Laboratory of Physical Geography and Environmental Process,Xining 810008,China;Academy of Plateau Science and Sustainability,Xining 810008,China)

机构地区：[1]青海师范大学地理科学学院,西宁810008 [2]青藏高原地表过程与生态保育教育部重点实验室,西宁810008 [3]青海省自然地理与环境过程重点实验室,西宁810008 [4]高原科学与可持续发展研究院,西宁810008

出　　处：《自然资源遥感》2022年第4期144-154,共11页Remote Sensing for Natural Resources

基　　金：青海省自然科学基金项目“基于GEE云平台与Landsat卫星长时间序列数据的湟水流域30多年土地利用/土地覆被时空变化研究”(编号:2021-ZJ-913)资助。

摘　　要：青海东部农业区地处黄土高原向青藏高原的过渡地带,黄土丘陵地貌类型多样、地形起伏大、破碎。随着近几十年来城市化进程的加快,农村可用劳动力缺失导致土地撂荒现象日益严重,因此掌握东部农业区撂荒地分布状况,对保护耕地与生态用地至关重要。本研究基于GEE云平台,以青海民和县为案例,依据农作物的物候特征,选取种植期和成熟期2季的Sentinel-2 MSI与Sentinel-1 SAR卫星影像为主要数据源,以DEM为辅助,结合光谱、地形、极化与缨帽特征,采用随机森林方法对研究区2018—2020年土地覆被进行自动分类,获取了研究区3 a的土地覆被数据,在此基础上借助撂荒地判断规则建立决策树提取撂荒地并进行验证。研究结果表明:2018年、2019年及2020年土地覆被总体分类精度分别为86.93%,87.36%和88.54%;2020年民和县撂荒地面积为43.17 km 2,占总面积的2.28%;撂荒地主要分布在海拔为2200~2600 m范围、坡度为6°~25°范围、坡向为阴坡的区域。Sentinel-1 SAR影像极化特征结合到Sentinel-2 MSI多季相数据中,能够有效提高黄土丘陵地形区土地覆被分类精度,获得较为准确的撂荒地信息。该研究为类似地形区域进行撂荒地提取提供了方法参考和借鉴。The eastern agricultural areas of Qinghai are located in the transitional zone from the Loess Plateau to the Qinghai-Tibet Plateau.In this transitional zone,the loess hills feature various landforms,large fluctuations,and fragmentation.With the acceleration of urbanization in recent decades,the shortage of available rural labor force has aggravated land abandonment.Therefore,ascertaining the distribution of abandoned land in the eastern agricultural areas of Qinghai is very crucial to protecting cultivated and ecological land.This study investigated Minhe County in Qinghai Province based on the GEE cloud platform.According to the phenological characteristics of crops,both Sentinel-2 MSI and Sentinel-1 SAR satellite images covering the growth and planting periods of crops were selected as the main data source.With the aid of the DEM and by combining the characteristics of spectra,terrain,polarization,and tasseled cap,this study automatically classified the land cover from 2018 to 2020 in the study area using the random forest method,obtaining the three-year land cover data of the study area.Then,this study built a decision tree based on the determination rules for abandoned land and extracted and verified the abandoned land information using the decision tree.The study results are as follows.The overall classification precision of land cover in 2018,2019,and 2020 were 86.93%,87.36%,and 88.54%,respectively.The area of abandoned land in Minhe County in 2020 was 43.17 km 2,accounting for 2.28% of the total study area.The abandoned land was mainly distributed in areas with an altitude of 2200~2600 m,a slope of 6°~25°,and a shady slope direction.The integration of the polarization characteristics of Sentinel-1 SAR images into Sentinel-2 MSI multi-season images can effectively improve the land cover classification precision and yield accurate information on the abandoned land.This study will provide a reference method and basis for the information extraction of abandoned land in areas with similar terrain.

关 键 词：撂荒地 Sentinel-1/2 多季相 随机森林 黄土高原西部 民和县 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]