Introducing recurrent event analyses to assess species interactions based on camera-trap data: A comparison with time-to-first-event approaches

Abstract

Camera-trap surveys are increasingly used to assess species interactions, with a growing focus on proximate co-occurrence, the conditional probability of one species occurring after another within a given time window. However, existing time-to-first-event models, commonly employed for this purpose, suffer from multiple limitations. For instance, they cannot fully quantify the interaction strength, that is how strongly one species affects the occurrence probability of another, nor the duration of this effect, nor identify the directionality of the interaction. We propose reframing the question in terms of recurrent event analysis, which enables the use of a rich set of techniques to provide more information about the data-generating process. In particular, we use the piece-wise exponential additive mixed model (PAMM) to take into account all available information (including recurrent occurrences of species) and to estimate the non-linear temporal dynamics of the visitation rate of a species after the occurrence of another. As the estimation is performed within the framework of generalized additive models, it offers a rich, robust and established inferential toolbox. We evaluated PAMM's performance in estimating species interactions through a simulation study involving two species under various scenarios of attraction or repulsion, with and without imperfect detection. Additionally, we compared the PAMM to six time-to-first-event methods commonly used in camera-trap studies and illustrated its application with real data. PAMMs outperformed the other methods in terms of statistical power, output interpretability and ability to disentangle interactions. The robust performance of PAMMs underscores their potential as a valuable tool for studying species interactions based on camera-trap data. To facilitate its adoption, we present the new package ctrecurrent and provide a reproducible workflow from data preparation to model fitting.