欢迎访问林草资源研究

林草资源研究 ›› 2024›› Issue (1): 65-72.doi: 10.13466/j.cnki.lczyyj.2024.01.009

• 科学研究 • 上一篇    下一篇

基于机器学习的红树林生物量遥感反演研究

郝君1,3(), 吕康婷2, 胡天祺4, 王云阁1,5, 徐刚1,6()   

  1. 1.浙江安防职业技术学院,浙江 温州 325000
    2.温州市天空地态势感知应用技术协同创新中心,浙江 温州 325000
    3.中国矿业大学,江苏 徐州 221116
    4.温州市未来城市研究院,浙江 温州 325000
    5.温州市自然灾害遥感监测重点实验室,浙江 温州 325000
    6.中南大学地球科学与信息物理学院,长沙 410083
  • 收稿日期:2023-10-16 修回日期:2024-01-12 出版日期:2024-02-28 发布日期:2024-03-22
  • 通讯作者: 徐刚,教授,主要研究方向:人工智能、地理空间信息技术、遥感智能解译。Email:xugang68@163.com
  • 作者简介:郝君,高级工程师,主要研究方向:自然资源信息化、空间大数据分析、遥感智能解译。Email:68502484@qq.com
  • 基金资助:
    温州市基础性科研项目“基于‘天空地’一体化及InVEST模型的红树林碳汇能力监测评估研究”(S2023030);浙江省教育厅一般科研项目“顾及物候知识的互花米草遥感监测方法的研究”(Y202351948)

Remote Sensing Inversion of Mangrove Biomass Based on Machine Learning

HAO Jun1,3(), LYU Kangting2, HU Tianqi4, WANG Yunge1,5, XU Gang1,6()   

  1. 1. Zhejiang College of Security Technology,Wenzhou 325000,Zhejiang,China
    2. Wenzhou Collaborative Innovation Center for Space-borne,Airborne and Ground Monitoring Situational Awareness Technology,Wenzhou 325000,Zhejiang,China
    3. China University of Mining and Technology,Xuzhou 221116,Jiangsu,China,4.Wenzhou Future City Research Institute,Wenzhou 325000,Zhejiang,China
    4. Wenzhou Key Laboratory of Natural Disaster Remote Sensing Monitoring and Early Warning,Wenzhou 325000,Zhejiang,China
    5. School of Geosciences and Info-Physics,Central South University,Changsha 410083,China
  • Received:2023-10-16 Revised:2024-01-12 Online:2024-02-28 Published:2024-03-22

摘要:

准确调查红树林生物量有利于评估红树林生态系统碳汇潜力。基于实地调查数据和Landsat 8遥感影像及DEM数据提取22个特征变量,利用随机森林(RF)、支持向量机模型(SVM)和极端梯度提升(XGBoost)3种机器学习方法,对西门岛红树林进行生物量遥感反演。结果表明:1)与RF算法和SVM算法相比,XGBoost算法构建的模型具有更好的估测效果(R2=0.932,ERMS=1.514 t/hm2,EMA=1.313 t/hm2),能更准确地估测红树林生物量;2)在递归特征消除法(RFE)筛选出的10个重要特征因子中,植被指数对红树林生物量估测的相对重要性较高;3)10个重要特征因子构成的XGBoost模型生物量反演得出,红树林生物量估测值范围为9.138~29.229 t/hm2,这与实地调查结果非常相近。XGBoost机器学习算法在红树林生物量反演中表现出较好的效果,该结果能为中国人工红树林碳储量的核算提供技术参考。

关键词: 红树林, Landsat 8, XGBoost, 生物量, 遥感反演

Abstract:

Accurately investigating mangrove biomass is beneficial for evaluating the carbon sink potential of mangrove ecosystems.Based on field survey data,Landsat 8 remote sensing images and DEM data,22 feature variables were extracted to carry out remote sensing inversion of mangrove biomass in the Ximen Island,which used three machine learning methods:Random Forest(RF),Support Vector Machine model(SVM)and eXtreme Gradient Boosting(XGBoost).The results showed:1)Compared to the RF model and SVM model,the XGBoost model had a better estimation performance(R2=0.932,ERMS=0.514 t/hm2,EMA=0.313 t/hm2),which could more accurately estimate the mangrove biomass.2)Among the 10 important characteristic factors selected by Recursive Feature Elimination(RFE),the vegetation index has a relatively high importance in estimating mangrove biomass.3)The biomass inversion map of the XGBoost model,which is composed of 10 important characteristic factors,showed that the estimated mangrove biomass ranges from 9.138 to 29.229 t/hm2,which was similar to the findings of the field survey.It can be seen that the XGBoost algorithm shows good application capabilities in mangrove biomass.This research will provide a technical reference for the accounting of carbon storage in the Chinese mangroves.

Key words: mangroves, Landsat 8, XGBoost, biomass, remote sensing inversion.

中图分类号: