How to account for behavioural states in step-selection analysis: a model comparison

TL;DR

This paper introduces a new Markov-switching integrated step-selection analysis (MS-iSSA) that jointly estimates animal behavioral states and habitat preferences, improving accuracy over standard methods that ignore behavioral states.

Contribution

The paper develops and evaluates MS-iSSA, a novel modeling framework combining hidden Markov models with step-selection analysis, and demonstrates its advantages over existing approaches.

Findings

Standard iSSA can produce biased estimates when ignoring behavioral states.

Prior state classification approaches ignore uncertainty, leading to errors.

MS-iSSA accurately estimates behavioral states and habitat preferences.

Abstract

Step-selection models are widely used to study animals' fine-scale habitat selection based on movement data. Resource preferences and movement patterns, however, can depend on the animal's unobserved behavioural states, such as resting or foraging. This is ignored in standard (integrated) step-selection analyses (SSA, iSSA). While different approaches have emerged to account for such states in the analysis, the performance of such approaches and the consequences of ignoring the states in the analysis have rarely been quantified. We evaluated the recent idea of combining hidden Markov chains and iSSA in a single modelling framework. The resulting Markov-switching integrated step-selection analysis (MS-iSSA) allows for a joint estimation of both the underlying behavioural states and the associated state-dependent habitat selection. In an extensive simulation study, we compared the MS-iSSA to both the standard iSSA and a classification-based iSSA (i.e., a two-step approach based on a separate prior state classification). We further illustrate the three approaches in a case study on fine-scale interactions of simultaneously tracked bank voles (Myodes glareolus). The results indicate that standard iSSAs can lead to erroneous conclusions due to both biased estimates and unreliable p-values when ignoring underlying behavioural states. We found the same for iSSAs based on prior state-classifications, as they ignore misclassifications and classification uncertainties. The MS-iSSA, on the other hand, performed well in parameter estimation and decoding of behavioural states. To facilitate its use, we implemented the MS-iSSA approach in the R package msissa available on GitHub.