Data from: A multispecies occupancy model for two or more interacting species

Species occurrence is influenced by environmental conditions and the presence of other species. Current approaches for multispecies occupancy modelling are practically limited to two interacting species and often require the assumption of asymmetric interactions. We propose a multispecies occupancy model that can accommodate two or more interacting species. We generalize the single-species occupancy model to two or more interacting species by assuming the latent occupancy state is a multivariate Bernoulli random variable. We propose modelling the probability of each potential latent occupancy state with both a multinomial logit and a multinomial probit model and present details of a Gibbs sampler for the latter. As an example, we model co-occurrence probabilities of bobcat (Lynx rufus), coyote (Canis latrans), grey fox (Urocyon cinereoargenteus) and red fox (Vulpes vulpes) as a function of human disturbance variables throughout 6 Mid-Atlantic states in the eastern United States. We found evidence for pairwise interactions among most species, and the probability of some pairs of species occupying the same site varied along environmental gradients; for example, occupancy probabilities of coyote and grey fox were independent at sites with little human disturbance, but these two species were more likely to occur together at sites with high human disturbance. Ecological communities are composed of multiple interacting species. Our proposed method improves our ability to draw inference from such communities by permitting modelling of detection/non-detection data from an arbitrary number of species, without assuming asymmetric interactions. Additionally, our proposed method permits modelling the probability two or more species occur together as a function of environmental variables. These advancements represent an important improvement in our ability to draw community-level inference from multiple interacting species that are subject to imperfect detection.物种出现情况受环境条件以及其他物种的存在状况影响。当前的多物种占有建模方法在实际应用中通常仅可处理两个互动物种，且往往需要假设物种间的交互作用为非对称形式。 我们提出了一种可适配两个及以上互动物种的多物种占有模型。通过假设潜在占有状态为多元伯努利（multivariate Bernoulli）随机变量，我们将单物种占有模型推广至多物种交互场景。我们提出采用多项logit模型与多项probit模型对每一种潜在占有状态的概率进行建模，并详述了后者（多项probit模型）对应的吉布斯采样器（Gibbs sampler）实现细节。 作为示例，我们以美国东部大西洋沿岸中部的6个州为研究范围，以人类干扰变量为协变量，构建了山猫（Lynx rufus）、郊狼（Canis latrans）、灰狐（Urocyon cinereoargenteus）以及赤狐（Vulpes vulpes）的共现概率模型。研究发现多数物种间存在成对交互作用的证据，且部分物种种对的同站点占有概率会随环境梯度发生变化：例如，在人类干扰程度较低的样点，郊狼与灰狐的占有概率相互独立；而在人类干扰程度较高的样点，这两个物种同时出现的概率显著升高。 生态群落由多个相互作用的物种构成。我们提出的方法允许对任意数量物种的检测/未检测数据进行建模，且无需假设交互作用为非对称形式，从而提升了针对这类群落开展统计推断的能力。此外，该方法还支持以环境变量为协变量，对两个及以上物种的共现概率进行建模。上述改进显著提升了我们从存在不完全检测（imperfect detection）的多个互动物种中获取群落层面统计推断的能力。