STATISTICAL CLASSIFICATION OF DIVING BEHAVIOR
STATISTICAL CLASSIFICATION OF DIVING BEHAVIOR
复制标题
潜水行为统计分类
DOI：
10.1111/j.1748-7692.1995.tb00277.x
复制
发表时间：
1995
影响因子：
2.3
通讯作者：
J. Ward Testa
中科院分区：
文献类型：
--
作者：
J. F. Schreer;J. Ward Testa
潜水行为的研究直接通过观察动物，间接通过分析数据，代表行为（如，潜水深度）。最新的研究利用数据从时间深度记录器（TDR）主要分组潜水行为主观根据感知的相似性，在最大深度，持续时间，和一般外观的潜水剖面（深度VJ。时间）（Kooyman 1968，Le Boeuf et ul. 1988，DeLong和Stewart 1991，Goebel等1991，Hindell等1991，Bengtson和Stewart 1992）。由于这些设备记录的数据量巨大，因此需要更有效和更客观的分析类型。行为的统计分析可能有助于加快大型数据集的分析，在数据压缩是必要的（例如，卫星传输）提供“板”分类，并减少人类主观偏见解释潜水行为。威德尔海豹（Leptonychotes weddellii）是检验多变量统计技术的一个很好的模型，因为已经收集了大量的潜水数据，而且它们的潜水行为相对来说是众所周知的和多样的（Kooyman 1968，1975，1981; Kooyman等人1983; Testa等人1989; Castellini等人1992; Testa，出版中）。威德尔海豹潜水最初根据潜水的最大深度和持续时间分为三种类型（Kooyman 1968）。本文描述了一种对威德尔海豹潜水大数据集进行统计分类的方法，并对三种不同类型的用于分组观测的多元技术进行了测试，以确定哪种技术最适合对威德尔海豹潜水进行分组：主成分分析、判别函数分析和聚类分析。主成分分析检查几个定量变量之间的关系，并得出这些变量的线性组合（主成分），这些变量保留了原始数据中的尽可能多的信息。
Diving behavior has been studied directly by observing animals, and indirectly by analyzing data that represents the behavior (eg, dive depths). Most recent studies utilizing data from time-depth recorders (TDRs) have primarily grouped diving behavior subjectively according to perceived similarities in the maximum depth, duration, and general appearance of the dive profile (depth VJ. time)(Kooyman 1968, Le Boeuf et ul. 1988, DeLong and Stewart 1991, Goebel et al. 1991, Hindell et al. 1991, Bengtson and Stewart 1992). The need for more efficient and objective types of analyses has developed due to the enormous amount of data recorded by these devices. Statistical analyses of behavior may be useful in expediting the analysis of large data sets, providing “ board” classification where data compression is necessary (eg, satellite transmissions), and reducing human subjective bias in interpreting diving behavior. Weddell seals (Leptonychotes weddellii) are a good model on which to test multivariate statistical techniques because large amounts of dive data have been collected and their diving behavior is relatively well known and diverse (Kooyman 1968, 1975, 1981; Kooyman et al. 1983; Testa et al. 1989; Castellini et al. 1992; Testa, in press). Weddell seal dives were originally classified into three types depending on the maximum depth and duration of the dives (Kooyman 1968). In this paper an approach to statistically classify large data sets of Weddell seal dives is described.Three different types of multivariate techniques used in grouping observations were tested to see which would be the most appropriate for grouping Weddell seal dives: principal component analysis, discriminant function analysis, and cluster analysis. Principal component analysis examines relationships among several quantitative variables and derives linear combinations(principal components) of these variables that retain as much of the information in the original