欢迎访问《草业学报》官方网站,今天是 分享到:

草业学报 ›› 2021, Vol. 30 ›› Issue (4): 1-12.DOI: 10.11686/cyxb2020190

• 研究论文 • 上一篇    下一篇

基于XGBoost算法的草甸地上生物量的高光谱遥感反演

张亦然1(), 刘廷玺1,2(), 童新1,2, 段利民1,2, 吴宇辰1   

  1. 1.内蒙古农业大学水利与土木建筑工程学院,内蒙古 呼和浩特 010018
    2.内蒙古自治区水资源保护与利用重点实验室,内蒙古 呼和浩特 010018
  • 收稿日期:2020-04-28 修回日期:2020-06-16 出版日期:2021-04-20 发布日期:2021-03-16
  • 通讯作者: 刘廷玺
  • 作者简介:Corresponding author. E-mail: txliu1966@163.com
    张亦然(1996-),女,内蒙古巴彦淖尔人,在读硕士。E-mail: 534805685@qq.com
  • 基金资助:
    国家自然科学基金项目(51620105003);内蒙古自然科学基金项目(2018ZD05);教育部创新团队发展计划(IRT_17R60);科技部重点领域科技创新团队(2015RA4013);内蒙古农业大学高层次人才科研启动金项目(NDYB2017-24);内蒙古自治区草原英才产业创新创业人才团队,内蒙古农业大学寒旱区水资源利用创新团队(NDTD2010-6);内蒙古农业大学学生科技创新基金项目(KJCX2019019)

Hyperspectral remote sensing inversion of meadow aboveground biomass based on an XGBoost algorithm

Yi-ran ZHANG1(), Ting-xi LIU1,2(), Xin TONG1,2, Li-min DUAN1,2, Yu-chen WU1   

  1. 1.College of Water Conservancy and Civil Engineering,Inner Mongolia Agricultural University,Hohhot 010018,China
    2.Inner Mongolia Water Resource Protection and Utilization Key Laboratory,Hohhot 010018,China
  • Received:2020-04-28 Revised:2020-06-16 Online:2021-04-20 Published:2021-03-16
  • Contact: Ting-xi LIU

摘要:

准确、高效获取草甸地上生物量信息,对牧区农牧业生产、草地资源管理、牧草可持续利用具有重要意义。本研究基于实地采集的牧草冠层光谱反射率及同期获取的地上生物量数据,运用互信息法分别分析了微分光谱、优化植被指数与草甸地上生物量的相关性,进一步构建了极限梯度提升(XGBoost)算法与不同阶光谱植被指数数据集的草甸地上生物量模拟估算模型,并与多元线性回归(MLR)和随机森林(RF)算法建立的模型进行对比。结果表明:对光谱反射率进行一阶、二阶微分与光谱植被指数变换协同应用,有助于提高冠层光谱与地上生物量的相关性;基于原始光谱植被指数与XGBoost算法构建的草甸地上生物量模拟估算模型效果最佳,均方根误差(RMSE)为140.26 g·m-2,平均绝对误差(MAE)为97.20 g·m-2,Nash效率系数(NSE)为0.81,一致性指数(d)为0.94,其次为基于RF算法构建的模型,MLR算法构建的模型精度较差。研究认为XGBoost算法可适用于草甸地上生物量模拟估算模型的建立,为快速准确的牧草高光谱遥感监测提供了技术和方法,为区域性草地高精度大面积生产力估算奠定了基础。

关键词: 地上生物量, XGBoost算法, 高光谱反射率, 植被指数

Abstract:

It is of great significance to agricultural production and animal husbandry, the management of grassland resources, and the sustainable utilization of forage to have efficient access to accurate information on aboveground biomass. This paper uses field data on the spectral reflectance of the grass canopy together with data on the aboveground biomass obtained in the same period, to analyze the correlation between spectral differences and meadow aboveground biomass, using a mutual information method. The correlation between the optimized vegetation index and the aboveground biomass of the meadow is then analyzed. In this paper, an extreme gradient boosting (XGBoost) algorithm and a simulation model estimating the aboveground biomass of different vegetation spectral index data sets are constructed. This model is compared with a model established by multiple linear regression (MLR) and random forest (RF) methodology. It was found that the first-order and second-order differentiation of spectral reflectance curves and the conjoint application of spectral vegetation index transformation from both spectral differentiation orders is helpful to improve the correlation between canopy spectra and aboveground biomass. Of the three approaches tested using the original spectral vegetation index to predict aboveground biomass, the XGBoost algorithm simulation model performed best, followed by the RF model. The MLR model displayed poor accuracy. The root mean square error of the XGBoost model was 140.26 g·m-2; the mean absolute error was 97.20 g·m-2; the Nash-Sutcliffe efficiency coefficient was 0.81, and the index of agreement was 0.94. It is concluded that the XGBoost algorithm can be used to create models for the estimation of steppe meadow aboveground biomass. The technology developed in this study provides a method for the rapid and accurate hyperspectral remote sensing monitoring of forage biomass, and lays a foundation for high-precision estimation of regional grassland herbage accumulation on a regional scale.

Key words: aboveground biomass, extreme gradient boosting algorithm, hyperspectral reflectance, vegetation index