草地学报 ›› 2023, Vol. 31 ›› Issue (7): 1964-1976.DOI: 10.11733/j.issn.1007-0435.2023.07.006

• 研究论文 • 上一篇    下一篇

黄河源永曲河流域高寒草甸地上生物量模拟与时空分布特征研究

李希来1, Jay Gao2, 师研2   

  1. 1. 青海大学农牧学院, 青海 西宁 810016;
    2. 奥克兰大学环境学院, 新西兰 奥克兰 1010
  • 收稿日期:2022-12-28 修回日期:2023-03-16 出版日期:2023-07-15 发布日期:2023-08-01
  • 通讯作者: 师研,E-mail:yshi917@aucklanduni.ac.Nz
  • 作者简介:李希来(1964-),男,河南孟津县人,教授,主要从事高寒草地生态环境模拟、生态修复和管理的教学和科学研究,E-mail:xilai-li@163.com
  • 基金资助:
    青海省自然科学基金创新团队项目(2020-ZJ-904);国家自然科学联合基金项目(U21A20191);高等学校学科创新引智计划项目(D18013) ;青海省科技创新创业团队项目“三江源生态演变与管理创新团队”资助

Aboveground Biomass Simulation and Its Temporal-Spatial Variation of Yongqu River Basin in the Alpine Meadow in the Yellow River Source Zone

LI Xi-lai1, GAO Jay2, SHI Yan2   

  1. 1. College of Agriculture and Animal Husbandry, Qinghai University, Xining, Qinghai Province 810016, China;
    2. School of Environment, the University of Auckland, Auckland 1010, New Zealand
  • Received:2022-12-28 Revised:2023-03-16 Online:2023-07-15 Published:2023-08-01

摘要: 本研究基于谷歌引擎通过四种常用模型及多种输入组合(地理空间变量(Geospatial variables,GV),植物功能类型(Plant functional types,VT),地面测量(Ground measurements,GM),气象变量(Meteorological variables,MV))对黄河源区高寒草甸地上生物量(Aboveground biomass,AGB)进行了模拟,并分析了AGB的时空分布与地形因子的关系。结果表明,仅使用GV构建的模型表现较差(0.122<R2<0.486),MV和VT分别与GV结合使用时能提高模拟精度0.104~0.203(R2),GM与GV结合使用时,模型精度达到了最高(0.678<R2<0.705)。在没有GM参与的情况下,深度神经网络(Deep neural network,DNN)模型结合GV-VT-MV变量组合获得了最好模拟精度为0.686(R2)。混合使用多种植被类型的数据可以提高模拟精度。本研究发现海拔是影响黄河源流域单位内高寒草甸AGB时空分布的重要决定因素,并且对AGB年变化量影响最强。

关键词: 青藏高原高寒草甸, 机器学习, 谷歌引擎, 地上生物量模拟, 地上生物量空间分布

Abstract: It is critical to model and map alpine meadow aboveground biomass (AGB) accurately for pastoral sustainable management on the Qinghai-Tibet Plateau. This study evaluated the performance of the four models and various input features combinations for grassland AGB modelling and mapping based on Google Earth Engine in the Yellow River source zone. The former includes the traditional multiple linear regression (MLR),support vector machine (SVM),artificial neural network (ANN),and deep neural network (DNN),the latter the ground measurements (GM),geospatial variables (GV),meteorological variables (MV),plant functional types (VT). The results showed that the solely use of GV had poor performance in AGB simulation (0.122<R2<0.486). The involvement of MV and VT into GV improved the accuracy (R2) by 0.104~0.203. The combination of GM-GV improved the accuracy to the highest level (0.678<R2<0.705),but models with GM has limitation to map AGB. However,without GM,DNN achieved the highest accuracy of 0.686 (R2) using the feature combination of GV-VT-MV. It was found out that the use of multiple vegetation dataset improved the AGB estimation accuracy. Further analysis between the temporal-spatial AGB distribution and topographic factors showed that elevation is the most important determinant of AGB distribution,and it also has the strongest influence on the variation of annual AGB. The method proposed in this study was able to model and map alpine meadow AGB and its variation,it could be used in the management of the alpine meadow grassland for sustainable development.

Key words: Qinghai Tibet Plateau alpine meadow, Machine learning, Google Earth Engine, Aboveground biomass simulation, Spatial distribution of aboveground biomass

中图分类号: