Acta Agrestia Sinica ›› 2021, Vol. 29 ›› Issue (5): 946-955.DOI: 10.11733/j.issn.1007-0435.2021.05.011

Previous Articles     Next Articles

Research on Remote Sensing Estimation of Forage Above-ground Biomass Based on Optimal Model Selection

GUO Chao-fan, CHEN Ze-wei, ZHANG Zhi-gao   

  1. College of Resources Environment and Tourism, Anyang Normal University, Anyang, Henan Province 455000, China
  • Received:2020-10-29 Revised:2021-01-20 Online:2021-05-15 Published:2021-06-02

基于最优模型选择的牧草地上生物量遥感估算研究

郭超凡, 陈泽威, 张志高   

  1. 安阳师范学院资源环境与旅游学院, 河南 安阳 455000
  • 通讯作者: 郭超凡,E-mail:guochao881016@163.com
  • 作者简介:郭超凡(1988-),男,汉族,山西吕梁人,博士,讲师,主要从事植被定量遥感研究,E-mail:guochao881016@163.com
  • 基金资助:
    国家自然科学基金项目(41602366);河南省高等学校重点科研项目(21A170004);安阳市科技计划项目(2021C01SF021);大学生创新创业训练计划项目(X2020104790155)共同资助

Abstract: Above-ground biomass of grassland is one of the key indicators for evaluating the grassland ecosystem productivity,and accurate estimation of forage biomass has important guiding significance for the real-time monitoring of growth status,the sustainable use,and management of grassland resources. The sentinel-2 image and ground measured data of the study area,which was located in Haiyan country,Qinghai Province,were used for the estimation of forage biomass to explore the optimal model. To select the optimal regression model,21 typical relationship models related to vegetation index and forage biomass data were constructed,which included a univariate index model,multiple linear regression model,and random forest model. Moreover,the model accuracy was tested through the Leave-one-out cross-validated coefficient of determination (Rcv2) and the cross-validated root mean square error (RMSEcv). Finally,the optimal inversion model was used to map the spatial distribution of biomass of the study area. The results showed that CIgreenand NDWI had better performance in fitting relationship with forage biomass. The CIre,MTVI2,GVMI2,and WDRVI,which indicate the water content and chlorophyll content of forage,also had important contributions in the multivariate model,which indicated that the growth and water content of forage had a great influence on the estimation of forage wet biomass. Besides,the accuracy of estimated results could be effectively improved by eliminating image noise (atmospheric noise and background noise). In the three inversion models,the random forest model had the best estimation accuracy,Rcv2reached 0.74,and RMSEcv reached 187.71 g·m-2. In conclusion,the study used Sentinel-2 data combining with random forest algorithm to form an accurate estimation of forage biomass. The results could provide a theoretical basis for remote sensing monitoring of forage biomass in the future and a reference for sustainable development and utilization of grassland.

Key words: Forage, Biomass estimation, Sentinel-2 image, Random forest, Grassland in Qinghai Province

摘要: 生物量快速精准监测对于草地资源的可持续开发和利用具有重要意义。以青海省海晏县草场为研究区,采用Sentinel-2影像结合地面实测数据进行牧草生物量估算研究并探究最优模型。构建21种典型植被指数与生物量的关系模型,包括单变量指数模型、多元线性模型和随机森林模型,并采用留一交叉验证决定系数和均方根误差进行模型精度检验和最优模型选择。结果表明:绿色叶绿素指数(Green chlorophyll index,CIgreen)和归一化差异水体指数(Normalized difference water index,NDWI)与生物量具有最优拟合关系,同样反映牧草水分和叶绿素含量的其他指数在模型中也具有重要的贡献占比,说明牧草的长势和水分含量对牧草生物量的估算具有较大影响。3种模型中,多元模型精度高于单变量模型,非线性模型精度高于线性模型。其中随机森林模型效果最优,Rcv2达到0.74,RMSEcv达到187.71 g·m-2,并且该模型在缓解“过饱和”方面也具有明显优势。

关键词: 牧草, 生物量估算, Sentinel-2影像, 随机森林, 青海省草地

CLC Number: