A penalized likelihood for multispecies occupancy models improves predictions of species interactions

Abstract

Multispecies occupancy models estimate dependence among multiple species of interest from patterns of co‐occurrence, but problems associated with separation and boundary estimates can lead to unreasonably large estimates of parameters and associated standard errors when species are rarely observed at the same site or when data are sparse. In this paper, we overcome these issues by implementing a penalized likelihood, which introduces a small bias in parameter estimates in exchange for a potentially large reduction in variance. We compare parameter estimates obtained from both penalized and unpenalized multispecies occupancy models fit to simulated data that exhibit various degrees of separation and to a real‐word data set of bird surveys with little apparent overlap between potentially interacting species. Our simulation results demonstrate that penalized multispecies occupancy models did not exhibit boundary estimates and produced lower bias, lower mean squared error, and improved inference relative to unpenalized models. When applied to real‐world data, our penalized multispecies occupancy model constrained boundary estimates and allowed for meaningful inference related to the interactions of two species of conservation concern. To facilitate the use of our penalized multispecies occupancy model, the techniques demonstrated in this paper have been integrated into the unmarked package in R programing language.

Using integrated multispecies occupancy models to map co-occurrence between bottlenose dolphins and fisheries in the Gulf of Lion, French Mediterranean Sea

“…While multispecies occupancy models require substantial amount of data to precisely estimate cooccurrence (Clipp et al 2021), integrated approaches can provide stronger inferences compared to an analysis of each dataset in isolation (Zipkin et al 2019. Our integrated multispecies occupancy model helped to overcome data scarcity and produced more precise estimations of co-occurrence probabilities than multispecies models using separated datasets (Supplementary materials).…”

“…To account for these issues, multispecies occupancy models have been developed to estimate occupancy probabilities of two or more interacting species while accounting for imperfect detection (Rota et al 2016b, Fidino et al 2019. One caveat of multispecies models is that they require substantial data for robust ecological inference (Clipp et al 2021). To overcome data scarcity, several authors have suggested to combine multiple datasets into an integrated modelling framework (see Kéry & Royle (2020), Chapter 10, for a review).…”

A framework to interpret co‐occurrence patterns from camera trap data: The case of the gray fox, the bobcat, and the eastern cottontail rabbit in a tropical dry habitat

“…Despite this, for our study, the existence of other important variables or interacting species influencing the occurrence of bobcats, gray fox, or cottontail rabbits is unlikely, since their omission would have been reflected in a substantial lack of goodness of fit of the occupancy models (additional variation in the data that is not explained by the model (Kéry & Royle, 2015)). Moreover, the inclusion of additional species in the analyses might have affected the precision and performance of the occupancy models (Rota et al., 2016) because multispecies occupancy models require a large amount of information (sites and records) to correctly estimate conditional occupancy parameters (Clipp et al., 2021; Kéry & Royle, 2021). The sensitivity of multispecies occupancy models to sample size and the number of records has not been formally tested and therefore requires more studies to check model performance and ability to detect co‐occurrence patterns under different sampling scenarios.…”

Development and validation of an eDNA protocol for monitoring endemic Asian spiny frogs in the Himalayan region of Pakistan

“…For both data types, initial modeling revealed problems with boundary estimates 91 . To address this, we fit all models using a penalized likelihood approach 91 , 92 .…”