检索
E-mail
RSS

作者投稿 主编审稿 专家审稿 远程编辑

热带海洋学报 ›› 2015, Vol. 34 ›› Issue (4): 37-47.doi: 10.11978/j.issn.1009-5470.2015.04.005CSTR: 32234.14.j.issn.1009-5470.2015.04.005

• 海洋物理学 • 上一篇    下一篇

基于支持向量机模型反演浮游植物粒级结构

姚林杰1, 2, 曹文熙1, 王桂芬1, 许占堂1, 胡水波1, 2, 周雯1, 李彩1   

  1. 1. 热带海洋环境国家重点实验室(中国科学院南海海洋研究所), 广东广州 510301;
    2. 中国科学院大学, 北京 100049;
  • 收稿日期:2014-10-02 出版日期:2015-08-10 发布日期:2015-08-21
  • 通讯作者: 曹文熙。E-mail: wxcao@scsio.ac.cn
  • 作者简介:姚林杰(1989~), 男, 广东省汕头市人, 硕士, 主要从事海洋环境监测。E-mail: linjiey@yahoo.com
  • 基金资助:

    国家自然科学基金 (41376042、41176035、41176083); 中国科学院海洋专项(XDA11040302); 国家自然科学基金青年基金项目(41206029); 中国科学院南海海洋研究所青年人才领域前沿项目(SQ201102); 热带海洋环境国家重点实验室(中国科学院南海海洋研究所)自主研究项目(LTOZZ1201)

A support vector machine model to estimate phytoplankton size classes

YAO Lin-jie1, 2, CAO Wen-xi1, WANG Gui-fen1, XU Zhan-tang1, HU Shui-bo1, 2, ZHOU Wen1, LI Cai1   

  1. 1.State Key Laboratory of Tropical Oceanography (South China Sea Institute of Oceanology, Chinese Academy of Sciences), Guangzhou 510301, China;
    2. University of Chinese academy of sciences, Beijing 100049, China;
  • Received:2014-10-02 Online:2015-08-10 Published:2015-08-21

PDF (PC)

赞

摘要：

文章采用支持向量机模型反演浮游植物的粒级结构。模型的输入量为浮游植物的吸收光谱、总叶绿素a浓度值。将该模型分别应用于南海数据集和NASA bio-Optical Marine Algorithm Dataset (NOMAD)全球大洋数据集。以浮游植物的吸收光谱做为输入向量时, 南海数据集和NOMAD数据集反演微微型(pico)、微型(nano)和小型(micro)粒级浮游植物的平均绝对误差(APD)分别是46.1%、61.6%、55.0%和36.3%、44.6%、43.3%; 决定系数(R2)分别为0.604、0.423、0.491和0.460、0.702、0.829。以浮游植物的吸收光谱和总叶绿素a浓度值做为输入向量时, 南海数据集和NOMAD数据集反演pico、nano和micro粒级的平均绝对误差分别是19.2%、31.9%、31.6%和35.3%、35.4%、38.2%; 决定系数分别为0.837、0.805、0.600和0.713、0.758、0.810。结果显示以吸收光谱和总叶绿素a浓度值作为输入变量的反演精度,比以吸收光谱作为输入变量的反演精度高。由此看出支持向量机模型对于两个数据集的反演结果很理想, 该模型的提出为多光谱遥感反演浮游植物的粒级结构提供一个重要手段。

关键词: 吸收光谱, 浮游植物粒级结构, 支持向量机, 叶绿素a

Abstract:

In this study, a support vector machine (SVM) model was introduced to retrieve phytoplankton size classes (PSCs), from phytoplankton absorption spectra and total chlorophyll-a concentration.The performance of this model was validated with the South China Sea andNASA bio-Optical Marine Algorithm Dataset(NOMAD) datasets. The results of the model, which used phytoplankton absorption spectra as the only input parameter, showed that the absolute percentage differences (APD) were 46.1% (pico), 61.6% (nano) and 36.3% (micro) for the South China Sea dataset, and were 36.3% (pico), 44.6% (nano) ,44.3% (micro) for NOMAD dataset; It also showed that the determination coefficents (R2) were 0.604 (pico), 0.423 (nano) and 0.491 (micro) for the South China Sea dataset, and were 0.460 (pico), 0.702 (nano) and 0.829 (micro) for the NOMAD dataset. The SVM model that used both absorption spectra and chlorophyll-a concentration of phytoplankton showed that the APD were 19.2%, 31.9%, 31.6% and determination coefficents of phytoplankton size classes (pico, nano, and micro) were 0.837, 0.805, 0.600 for the South China Sea data-set. Using the same method, the result of SVM model showed that the APD were 35.3%, 35.4%, 38.2% and determination coefficents of phytoplankton size classes (pico, nano, and micro) were 0.713, 0.758, 0.810 for the NOMAD data set. The SVM model trained using phytoplankton absorption spectra and total chlorophyll-a concentration performed more effectively than that using only phytoplankton absorption spectra. The performance of the SVM model was shown to be satisfactory, and the model opens the way to an appliaction to estimate PSCs by using hyperspectral measurements.

Key words: absorption spectra, phytoplankton size class, support vector machine, chlorophylls-a