基于Landsat-8和Sentinel-2时间序列合成影像的山区甘蔗种植区提取  

Extraction of Sugarcane Plantation in Mountainous Areas Based on Landsat-8 and Sentinel-2 Time-series Synthetic Images

作　　者：康倩文 徐伟恒 王雷光[2,3] 洪泽湖[1] 刘运 KANG Qianwen;XU Weiheng;WANG Leiguang;HONG Zehu;LIU Yun(College of Big Data and Intelligent Engineering,Southwest Forestry University,Kunming,Yunnan 650233,China;Institute of Big Data and Artificial Intelligence,Southwest Forestry University,Kunming,Yunnan 650233,China;Key Laboratory of Na-tional Forestry and Grassland Administration on Forestry and Ecological Big Data,Southwest Forestry University,Kunming,Yun-nan 650233,China)

机构地区：[1]西南林业大学大数据与智能工程学院,云南昆明650233 [2]西南林业大学大数据与智能工程研究院,云南昆明650233 [3]西南林业大学林业生态大数据国家林业与草原局重点实验室,云南昆明650233

出　　处：《热带作物学报》2023年第6期1276-1287,共12页Chinese Journal of Tropical Crops

基　　金：国家自然科学基金地区基金项目(No.32060320);云南省农业基础研究联合专项面上项目(No.202101BD070001-066);云南省重大科技专项(No.202102AE090051-1-02)。

摘　　要：山区耕地破碎化现象严重,种植结构复杂,加之山区多云雨天气使得遥感影像质量难以保证,给基于遥感的山区甘蔗种植区提取带来困难。为探究适用于山区甘蔗种植区提取的方法,本研究以典型山区甘蔗种植区云南省玉溪市新平县为研究区,以2019年10月1日至2021年7月1日过境新平县的Landsat-8和Sentinel-2光学影像为数据源,结合DEM数据和野外调查数据,基于谷歌地球引擎(google earth engine,GEE)云计算平台构建高空间、高时间分辨率的时间序列合成影像,借助甘蔗与常绿植被、水体、不透水层、其他农作物在光谱指数特征、物候特征、地形特征上的差异,采用上升时间、下降时间、上积分、下积分4个物候参数以及海拔、坡度因子确定提取甘蔗的最佳阈值,对研究区2020年甘蔗种植区进行提取并绘制2020年新平县甘蔗种植区分布图,最后对提取结果进行精度验证与甘蔗空间分布规律分析。结果表明:基于Landsat-8和Sentinel-2合成的时间序列影像可以增加研究区内像元的良好观测次数并提高影像的空间分辨率,克服了山区遥感影像质量不高的问题,可更好地监测植被物候特征与季节变化;本研究甘蔗种植区提取总体精度为97.07%,Kappa系数为0.83,其中甘蔗的用户精度为88.85%,制图精度为80.57%;新平县2020年甘蔗种植区面积为7705 hm^(2),甘蔗种植区的空间分布随新平县西北高东南低的地势呈现出东南多于西北的现象,且乡镇间甘蔗种植区面积差异显著,甘蔗种植面积最大的乡镇为漠沙镇(2786 hm^(2)),面积最小的乡镇为古城街道(0.87 hm^(2)),与实际调研情况一致。通过对物候参数进行敏感性分析发现,同时使用上升时间、下降时间、上积分、下积分4个物候参数进行甘蔗提取,可提高用户精度,减少错分误差。该研究提出的甘蔗种植区提取算法可为山区复杂地形下甘蔗种植区提取提供参考。Sugarcane plantations are mainly distributed in mountainous areas with high land fragmentation and complex cropping structure in Yunnan Province.Frequent cloud cover reduces the good observations of the land cover by using optical remote sensing.So it is difficult to extract sugarcane plantations with a high accuracy based on satellite optical remote sensing data.Xinping Country,a typical mountainous sugarcane plantation region,was chosen to explore a suitable method for the extraction of sugarcane plantations in mountainous areas.In this study,Landsat-8 and Sentinel-2 optical imagery for Xinping from October 1,2019 to July 1,2021 were used as the main data sources,and the DEM data and field survey data were used as the auxiliary data.The synthetic time-series images with high spatial-temporal resolution were constructed on Google Earth Engine(GEE).Firstly,we analyzed the differences among sugarcane and evergreen vegetation,water body,impervious,and other crops in spectral index characteristics,phenological characteristics,and topographic characteristics.Secondly,we determined the optimal thresholds for extracting sugarcane plantations for the four phenological parameters including rise time,fall time,above integral of season and below integral of season,as well as elevation and slope factors based on the training samples.Thirdly,we mapped the sugarcane plantations of 2020 and the mapping accuracy was verified using the validation samples in the study area.Finally,the spatial distribution of sugarcane plantations was analyzed at town scale.The results showed that synthetic time-series images based on the Landsat-8 and Sentinel-2 optical imagery could increase the number of good observations in the study area and improve the spatial resolution of the images,which could solve the problem of low quality of remote sensing images in mountainous areas and could better monitor phenological characteristics and seasonal changes of vegetation.The resultant 2020 sugarcane map had overall,user and producer accuracy of 97.07%

关 键 词：甘蔗 物候 时间序列影像 影像合成 谷歌地球引擎 

分 类 号：S566.1[农业科学—作物学]