Gimenez Olivier, Lebreton Jean-Dominique, Gaillard Jean-Michel, Choquet Rémi, Pradel Roger
Centre d'Ecologie Fonctionnelle et Evolutive, UMR 5175, CNRS, 1919 route de Mende, 34293 Montpellier Cedex 5, France.
Theor Popul Biol. 2012 Dec;82(4):307-16. doi: 10.1016/j.tpb.2012.02.001. Epub 2012 Feb 18.
Structured population models are widely used in plant and animal demographic studies to assess population dynamics. In matrix population models, populations are described with discrete classes of individuals (age, life history stage or size). To calibrate these models, longitudinal data are collected at the individual level to estimate demographic parameters. However, several sources of uncertainty can complicate parameter estimation, such as imperfect detection of individuals inherent to monitoring in the wild and uncertainty in assigning a state to an individual. Here, we show how recent statistical models can help overcome these issues. We focus on hidden process models that run two time series in parallel, one capturing the dynamics of the true states and the other consisting of observations arising from these underlying possibly unknown states. In a first case study, we illustrate hidden Markov models with an example of how to accommodate state uncertainty using Frequentist theory and maximum likelihood estimation. In a second case study, we illustrate state-space models with an example of how to estimate lifetime reproductive success despite imperfect detection, using a Bayesian framework and Markov Chain Monte Carlo simulation. Hidden process models are a promising tool as they allow population biologists to cope with process variation while simultaneously accounting for observation error.
结构化种群模型在植物和动物种群统计学研究中被广泛用于评估种群动态。在矩阵种群模型中，种群由个体的离散类别（年龄、生活史阶段或大小）来描述。为了校准这些模型，需要在个体层面收集纵向数据以估计种群统计学参数。然而，有几个不确定性来源会使参数估计变得复杂，比如野外监测中个体检测不完美以及给个体分配状态时的不确定性。在这里，我们展示了近期的统计模型如何有助于克服这些问题。我们聚焦于并行运行两个时间序列的隐藏过程模型，一个捕捉真实状态的动态，另一个由这些潜在的可能未知状态产生的观测值组成。在第一个案例研究中，我们通过一个使用频率论和最大似然估计来处理状态不确定性的例子来说明隐马尔可夫模型。在第二个案例研究中，我们通过一个使用贝叶斯框架和马尔可夫链蒙特卡罗模拟来估计尽管检测不完美但一生繁殖成功率的例子来说明状态空间模型。隐藏过程模型是一个很有前景的工具，因为它们允许种群生物学家在考虑观测误差的同时应对过程变化。
