测绘通报 ›› 2019, Vol. 0 ›› Issue (3): 46-52,75.doi: 10.13474/j.cnki.11-2246.2019.0076

• 学术研究 • 上一篇    下一篇

基于Google Earth Engine与机器学习的省级尺度零散分布草地生物量估算

修晓敏1,2, 周淑芳3, 陈黔1,2, 蒙继华2, 董文全2, 杨广斌1, 李晓松2   

  1. 1. 贵州师范大学地理与环境科学学院, 贵州 贵阳 550025;
    2. 中国科学院遥感与数字地球研究所 数字地球重点实验室, 北京 100094;
    3. 二十一世纪空间技术应用股份有限公司, 北京 100096
  • 收稿日期:2018-11-09 修回日期:2019-01-22 出版日期:2019-03-25 发布日期:2019-04-02
  • 通讯作者: 李晓松。E-mail:lixs@radi.ac.cn E-mail:lixs@radi.ac.cn
  • 作者简介:修晓敏(1993-),女,硕士生,主要从事生态遥感应用研究。E-mail:gznuxxm@126.com
  • 基金资助:
    高分辨率对地观测系统重大专项(30-Y20A03-9003-17/18);全省草地资源清查项目(2017FACZ2974)

Above-ground biomass estimation of provincial scattered grassland based on Google Earth Engine and machine learning

XIU Xiaomin1,2, ZHOU Shufang3, CHEN Qian1,2, MENG Jihua2, DONG Wenquan2, YANG Guangbin1, LI Xiaosong2   

  1. 1. School of Geography and Environmental Science, Guizhou Normal University, Guiyang 550025, China;
    2. Key Laboratory of Digital Earth Science, Institute of Remote Sensing and Digital Earth, Chinese Academy of Science, Beijing 100094, China;
    3. Twenty First Century Aerospace Technology Co., Ltd., Beijing 100096, China
  • Received:2018-11-09 Revised:2019-01-22 Online:2019-03-25 Published:2019-04-02

摘要: 大区域草地地上生物量估算对草地资源利用管理及全球碳循环研究具有重要意义。为高效快速地估算大区域零散分布草地地上生物量,本文选取安徽省为研究区,在谷歌地球云引擎(Google Earth Engine)平台的支撑下,通过机器学习方法建立Landsat 8 OLI及其他辅助数据与地面实测草地地上生物量之间的联系,开展了草地零散分布地区省级尺度草地地上生物量高分辨率估算,并与传统的基于归一化植被指数(NDVI)回归模型进行了比较。研究结果表明,综合利用光谱与地形因子的机器学习方法,估算零散化分布草地地上生物量的精度可以达到65%以上,其中分类回归树(CART)模型R2=0.57,预测精度为68.60%,支持向量机(SVM)模型R2=0.59,预测精度为75.74%,而使用NDVI的回归分析产生的误差较大,R2=0.37,预测精度为57.51%,因此机器学习方法相对于传统基于NDVI的回归分析具有明显优势。另外,谷歌地球云引擎平台数据来源广泛、获取方便,可以高效地实现海量影像数据的预处理及计算分析,大大提升了工作效率,与地面调查数据的结合可实现更大区域乃至全国尺度上的零散分布草地地上生物量高分辨率遥感估算。

关键词: 草地生物量, Google Earth Engine, 机器学习, 回归分析

Abstract: Estimating above-ground biomass of grassland in large areas is of great significance for grassland resource utilization,management and global carbon cycle research.In order to pursue efficient and rapid estimation of above-ground biomass of provincial scattered grassland,this study selected Anhui province as the research area,built the relationship between Landsat8 OLI,auxiliary data and measured above-ground biomass data through machine learning and Google Earth Engine (GEE) platform.The main results showed that the model which constructed by spectral information,terrain factors and machine learning had obvious advantages,the estimated accuracy was more than 65%.The classification and regression tree (CART) model R2 was 0.57,the estimated accuracy was 68.60%.Support vector machine (SVM) model R2 was 0.59,the estimated accuracy was 75.74%.The GEE platform has rich and available data,it can complete pre-processing and calculation analysis efficiently.The combination of GEE and ground survey data has the potential to estimate above-ground biomass of scattered grassland on a national scale.

Key words: grassland biomass, Google Earth Engine, machine learning, regression analysis

中图分类号: