University of Sheffield, Sheffield, UK.
Bielefeld University, Bielefeld, Germany.
Ecology. 2017 Jul;98(7):1932-1944. doi: 10.1002/ecy.1880. Epub 2017 Jun 12.
The behavior of colony-based marine predators is the focus of much research globally. Large telemetry and tracking data sets have been collected for this group of animals, and are accompanied by many empirical studies that seek to segment tracks in some useful way, as well as theoretical studies of optimal foraging strategies. However, relatively few studies have detailed statistical methods for inferring behaviors in central place foraging trips. In this paper we describe an approach based on hidden Markov models, which splits foraging trips into segments labeled as "outbound", "search", "forage", and "inbound". By structuring the hidden Markov model transition matrix appropriately, the model naturally handles the sequence of behaviors within a foraging trip. Additionally, by structuring the model in this way, we are able to develop realistic simulations from the fitted model. We demonstrate our approach on data from southern elephant seals (Mirounga leonina) tagged on Kerguelen Island in the Southern Ocean. We discuss the differences between our 4-state model and the widely used 2-state model, and the advantages and disadvantages of employing a more complex model.
基于群体的海洋捕食者的行为是全球许多研究的焦点。已经为这群动物收集了大量的远程遥测和跟踪数据集，并伴随着许多旨在以某种有用的方式对轨迹进行分段的实证研究，以及对最佳觅食策略的理论研究。然而，相对较少的研究详细介绍了推断中央觅食旅行中行为的统计方法。在本文中，我们描述了一种基于隐马尔可夫模型的方法，该方法将觅食旅行划分为标记为“出航”、“搜索”、“觅食”和“返航”的段。通过适当构造隐马尔可夫模型转移矩阵，该模型自然处理了觅食旅行中的行为序列。此外，通过以这种方式构建模型，我们能够从拟合的模型中开发出现实的模拟。我们在南大洋克尔格伦岛标记的南方象海豹（Mirounga leonina）的数据上展示了我们的方法。我们讨论了我们的 4 状态模型与广泛使用的 2 状态模型之间的差异，以及采用更复杂模型的优缺点。
