生物多样性 ›› 2017, Vol. 25 ›› Issue (9): 966-971.DOI: 10.17520/biods.2017019

• 研究报告: 植物多样性 • 上一篇    下一篇

用植物生活史性状预测种子扩散方式

郭志文, 郑景明*()   

  1. 北京林业大学林学院, 北京 100083
  • 收稿日期:2017-01-17 接受日期:2017-06-09 出版日期:2017-09-20 发布日期:2017-10-04
  • 通讯作者: 郑景明
  • 作者简介:# 共同第一作者
  • 基金资助:
    林业公益项目专项经费(201404303)

Predicting modes of seed dispersal using plant life history traits

Zhiwen Guo, Jingming Zheng*()   

  1. College of Forestry, Beijing Forestry University, Beijing 100083
  • Received:2017-01-17 Accepted:2017-06-09 Online:2017-09-20 Published:2017-10-04
  • Contact: Zheng Jingming
  • About author:# Co-first authors

摘要:

种子扩散方式对植物物种分布、种群动态及群落组成都有重要影响, 但目前有关种子扩散方式的数据还很欠缺。植物的生活史性状与种子扩散方式联系密切, 通过植物生活史性状预测种子的扩散方式是一种有效的研究手段。本文基于我国360种植物的生长型、株高、种子质量和果实类型以及种子扩散方式的数据集, 随机抽取288个物种数据(80%)作为训练样本, 采用神经网络、决策树、费舍尔线性判别和支持向量机算法, 分别建立种子扩散方式的预测模型, 将其余72个物种数据(20%)用于模型检验。以1,000次随机抽样后的平均判别正确率作为模型预测效果的评价指标。结果表明: 用生长型、株高、种子质量及果实类型作为主要预测变量, 构建的神经网络、决策树、费希尔线性判别和支持向量机模型均能达到较好的预测效果, 准确率分别为78.90%、77.09%、77.81%和78.14%, 其中以神经网络模型的预测效果最好。进一步研究发现, 神经网络模型对动物扩散、无助力扩散和风扩散的预测效果分别为81.32%、74.90%和81.45%。本研究为植物种子扩散方式预测提供了一种新的思路。

关键词: 扩散方式, 果实类型, 生长型, 株高, 种子质量, 预测模型

Abstract:

Mode of seed dispersal is an important trait for understanding geographical distributions, population dynamics, and community composition of plant species. However, data of dispersal modes are scarce for Chinese plant species. Previous studies have shown that growth form, plant height, fruit type, and seed mass have strong correlations with seed dispersal modes, thus predictions using modelling could be an alternative to gain this information. We collected information on growth forms, plant height, fruit types, seed mass, and dispersal modes from 360 kinds of Chinese angiosperm plants, and built a neural network model (NNET), decision tree (TREE), Fisher linear discriminant model (LDA), and support vector machine model (SVM) to predict seed syndromes from these four traits. For each model, an 80% sample (288 species) was randomly drawn from dataset as the training sample, with remaining 20% of data was used as a test sample. Results showed that all four models achieved rather good predictions, and the average total correctness rate for the NNET, TREE, LDA, and SVM was 78.90%, 77.09%, 77.81%, 78.14%, respectively. The neural network model had the highest correctness rates for different dispersal modes, i.e., zoochory (81.32%), autochory (74.90%), and anemochory (81.45%). This paper establishes the basis for the prediction of seed dispersal modes.

Key words: dispersal modes, fruit type, growth form, plant height, seed mass, prediction model