生物多样性 ›› 2018, Vol. 26 ›› Issue (8): 878-891.  DOI: 10.17520/biods.2018051

• • 上一篇    下一篇

基于激光雷达数据的物种分布模拟: 以美国加州内华达山脉南部区域食鱼貂分布模拟为例

周中一1, 刘冉1, 时书纳1, 苏艳军2, 李文楷1,*(), 郭庆华2,3   

  1. 1 中山大学地理科学与规划学院, 广州 510275
    2 中国科学院植物研究所植被与环境变化国家重点实验室, 北京 100093
    3 中国科学院大学, 北京 100049
  • 收稿日期:2018-02-10 接受日期:2018-04-28 出版日期:2018-08-20 发布日期:2018-09-27
  • 通讯作者: 李文楷
  • 作者简介:# 共同第一作者
  • 基金资助:
    国家自然科学基金(41401516)、广东省自然科学基金(2014A030313605)和广州市科技计划项目(2013J410006)

Ecological niche modeling with LiDAR data: A case study of modeling the distribution of fisher in the southern Sierra Nevada Mountains, California

Zhongyi Zhou1, Ran Liu1, Shuna Shi1, Yanjun Su2, Wenkai Li1,*(), Qinghua Guo2,3   

  1. 1 School of Geography and Planning, Sun Yat-Sen University, Guangzhou 510275
    2 State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, Beijing 100093
    3 University of Chinese Academy of Sciences, Beijing 100049
  • Received:2018-02-10 Accepted:2018-04-28 Online:2018-08-20 Published:2018-09-27
  • Contact: Li Wenkai
  • About author:# Co-first authors

摘要:

生态位模型通过拟合物种分布与环境变量之间的关系提供物种空间分布预测, 在生物多样性研究中有广泛应用。激光雷达(LiDAR)是一种新兴的主动遥感技术, 已被大量应用于森林三维结构信息的提取, 但其在物种分布模拟的应用研究比较缺乏。本研究以美国加州内华达山脉南部地区的食鱼貂(Martes pennanti)的分布模拟为例, 探索LiDAR技术在物种分布模拟中的有效性。生态位模型采用5种传统多类分类器, 包括神经网络、广义线性模型、广义可加模型、最大熵模型和多元自适应回归样条模型, 并使用正样本-背景学习(presence and background learning, PBL)算法进行模型校正; 同时对这5种模型使用加权平均进行模型集成, 作为第6个模型。此外, 一类最大熵模型也被用于模拟该物种的空间分布。模型的连续输出和二值输出分别使用AUC (area under the receiver operating characteristic curve)以及基于正样本-背景数据的评价指标Fpb进行评价。结果表明, 仅考虑气候因子(温度和降水)时, 7个模型的AUC和Fpb平均值分别为0.779和1.077; 当考虑LiDAR变量(冠层容重、枝下高、叶面积指数、高程、坡度等)后, AUC和Fpb分别为0.800和1.106。该研究表明, LiDAR数据能够提高食鱼貂空间分布的预测精度, 在物种分布模拟方面存在一定的应用价值。

关键词: 生态位模型, 激光雷达, 食鱼貂, 正样本-背景学习, 最大熵

Abstract

Ecological niche modeling seeks to infer the relationship between occurrences of a species and environmental covariates and has been widely applied in biodiversity studies. Light detection and ranging (LiDAR) is a new active remote sensing technology that is being increasingly used for acquisition of 3D structural information of forests. However, its applications in ecological niche modeling are rarely studied. In this study, we wanted investigate the effectiveness of LiDAR in modeling the spatial distribution of fisher (Martes pennanti) in the southern Sierra Nevada Mountains, California. We used artificial neural networks, generalized linear model, generalized additive model, discriminative maximum entropy, and multivariate adaptive regression splines to implement the presence and background learning (PBL) method separately. We then combined all the models based on weighted average to create an ensemble model. The generative maximum entropy model was also considered for comparison. Area under the receiver operating characteristic curve (AUC) and Fpb based on presence and background data were used to evaluate the continuous and binary outputs, respectively. Our results show that the values of AUC and Fpb were 0.779 and 1.077, respectively, when only climate variables (such as temperature and precipitation) were included in the models, whereas the values of AUC and Fpb were 0.800 and 1.106, respectively, when both climate and LiDAR-derived variables (such as canopy bulk density, height to live canopy base, leaf area index, digital elevation model, slope, etc.) were included in the models. Therefore, we conclude that LiDAR-derived variables are helpful in modeling the spatial distribution of fisher, and has good potential in ecological niche modeling.

Key words: ecological niche model, light detection and ranging (LiDAR), fisher, presence and background learning (PBL), maximum entropy (MaxEnt)