A multistate dynamic occupancy model to estimate local colonization–extinction rates and patterns of co‐occurrence between two or more interacting species

Abstract

Although ecology is rife with theory that explores how multiple species co‐occur through space and time, the field lacks robust statistical models to parameterize this theory with empirical data, particularly when species are detected imperfectly and data are collected as a time‐series. We address this need by developing an occupancy model that estimates local colonization and extinction rates for two or more interacting species when data are collected across multiple sampling occasions. This model estimates how community composition at a site may change across sampling occasions by assuming the latent occupancy state is a categorical random variable. We used a multinomial‐logit model to parameterize species specific parameters and pairwise interactions between species, both of which can be made a function of covariates. These transition probabilities between community states can then be converted to occupancy or co‐occurrence probabilities to determine how community composition varies along an environmental gradient or through time. As an example, we estimate patterns of co‐occurrence between coyote Canis latrans, Virginia opossum Didelphis virginiana, and raccoon Procyon lotor in Chicago, Illinois, USA with data from a multiyear camera trapping study. Models with pairwise interactions between species greatly out performed models that assumed independence between species. Opossum and raccoon, for example, were far less likely to go extinct in habitat patches where coyotes were present. Community composition at a site depends on species interactions and the local environment. Our model can separate such effects by estimating the underlying processes that define species occurrence patterns. As a result, our model can more explicitly quantify a wide range of ecological dynamics and therefore be used to empirically test ecological theory, such as estimating priority effects at a site or turnover rates between species, both of which can be made to vary as a function of covariates.

Inferring wildlife poaching in southeast Asia with multispecies dynamic occupancy models

“…To our knowledge, this study is one of the few applications of a two-species occupancy model extended to a multi-season version (see also Yackulick et al 2014, Fidino et al 2018. It is also the first to apply such a model to explicate the space use dynamics of a poacher-wildlife system.…”

“…Sites can then change from one state to another between successive primary occasions through extinction and colonisation processes. It also considered the probability that a species replaced another in a given site from one primary occasion to another (Fidino et al 2018). We adopted the Richmond-Waddle (RW) parameterization for occupancy and detection because it is more appropriate to describe poacher-wildlife interactions and identify the effects of covariates described below on species occupancy (Richmond et al 2010, Waddle et al 2010).…”

“…a site stays occupied by both from one year to another between t and t + 1 sake of simplicity, we described detection probabilities of a species (poachers or wildlife) not conditionally on the presence/absence of the other (Fidino et al 2018). We therefore defined the observation matrix as:…”

“…The strength of some of our inferences is limited because, unfortunately, we could not estimate the detection probability of wildlife only despite the presence of poachers and vice versa (Miller et al 2012, Fidino et al 2018. We ran preliminary analysis using the same model as in Miller et al (2012), where detection probability of one species was conditional on the presence of the other.…”

“…Using repeated surveys of detections and non-detections of poaching-related threats, such as snares, and of wildlife collected at several spatial units, one can estimate the temporal variation in the detection and occupancy probabilities of poachers in a given site conditional on the occurrence of wildlife with which they interact. Such a flexible parameterization can therefore assess the dynamics of species interactions simultaneously with species occupancy and detection (Fidino et al 2018). However, few studies account for the imperfect detection in the occupancy of poachers (Critchlow et al 2015, Moore et al 2018.…”