Accounting for behaviour in fine‐scale habitat selection: A case study highlighting methodological intricacies

Abstract

Animal habitat selection—central in both theoretical and applied ecology—may depend on behavioural motivations such as foraging, predator avoidance, and thermoregulation. Step‐selection functions (SSFs) enable assessment of fine‐scale habitat selection as a function of an animal's movement capacities and spatiotemporal variation in extrinsic conditions. If animal location data can be associated with behaviour, SSFs are an intuitive approach to quantify behaviour‐specific habitat selection. Fitting SSFs separately for distinct behavioural states helped to uncover state‐specific selection patterns. However, while the definition of the availability domain has been highlighted as the most critical aspect of SSFs, the influence of accounting for behaviour in the use‐availability design has not been quantified yet. Using a predator‐free population of high‐arctic muskoxen Ovibos moschatus as a case study, we aimed to evaluate how (1) defining behaviour‐specific availability domains, and/or (2) fitting separate behaviour‐specific models impacts (a) model structure, (b) estimated selection coefficients and (c) model predictive performance as opposed to behaviour‐unspecific approaches. To do so, we first applied hidden Markov models to infer different behavioural modes (resting, foraging, relocating) from hourly GPS positions (19 individuals, 153–1062 observation days/animal). Using SSFs, we then compared behaviour‐specific versus behaviour‐unspecific habitat selection in relation to terrain features, vegetation and snow conditions. Our results show that incorporating behaviour into the definition of the availability domain primarily impacts model structure (i.e. variable selection), whereas fitting separate behaviour‐specific models mainly influences selection strength. Behaviour‐specific availability domains improved predictive performance for foraging and relocating models (i.e. behaviours with medium to large spatial displacement), but decreased performance for resting models. Thus, even for a predator‐free population subject to only negligible interspecific competition and human disturbance we found that accounting for behaviour in SSFs impacted model structure, selection coefficients and predictive performance. Our results indicate that for robust inference, both a behaviour‐specific availability domain and behaviour‐specific model fitting should be explored, especially for populations where strong spatiotemporal selection trade‐offs are expected. This is particularly critical if wildlife habitat preferences are estimated to inform management and conservation initiatives.

Predicting fine-scale distributions and emergent spatiotemporal patterns from temporally dynamic step selection simulations

“…Another alternative method of incorporating temporal dynamics into animal movement models is through state-switching models such as a hidden Markov model (HMM) (Langrock et al, 2012; Mc-Clintock et al, 2012). In HMMs, behaviours such as foraging, resting and transiting are represented as states with different movement parameters, and when combined with SSFs (HMM-SSF/HMM-iSSA), different habitat selection parameters (Picardi et al, 2022; Beumer et al, 2023; Klappstein et al, 2023; Pohle et al, 2024). These models can easily incorporate temporal dynamics as the transition matrix governing state-switching can depend on time, although in HMMs the states are discrete, whereas real behaviour changes may be gradual and continuous, which may affect predictions in some cases.…”

“…Step selection functions (SSFs) are particularly advantageous as they can be used to simulate trajectories (Signer et al, 2017; Potts & Börger, 2023; Signer et al, 2023), are straightforward to parameterise, and can incorporate temporal dynamics (Ager et al, 2003; Forester et al, 2009; Tsalyuk et al, 2019; Richter et al, 2020; Klappstein et al, 2024). An SSF combines a movement and a external selection kernel, can take a range of forms (Munden et al, 2021; Klappstein et al, 2022; Beumer et al, 2023; Eisaguirre et al, 2024; Pohle et al, 2024), and can accommodate a wide range of covariates including habitat, linear features, distance-to-feature variables, proximity to other animals (Potts et al, 2022; Ellison et al, 2024) and representations of previous space use (Schlägel & Lewis, 2014; Oliveira-Santos et al, 2016; Rheault et al, 2021).…”

A new approach to geostatistical synthesis of historical records reveals capuchin spatial responses to climate and demographic change

“…The larger UD required for sleep-UDs to align with tracking-UDs, as well as the inability of sleep-UDs to recover the position of maximum space-use intensity, likely reflects that the areas capuchins use for sleeping have different characteristics than areas used for diurnal foraging. The concept that measures of space use are behaviourdependent is widely supported (Abrahms et al, 2016;Beumer et al, 2023;Klappstein et al, 2023;Séchaud et al, 2021;Suraci et al, 2019). Researchers should consider this when interpreting UDs from location records tied to one behavioural state.…”

Understanding step selection analysis through numerical integration

“…Note that the Monte Carlo approach described in Section 2.2.1 (where movement is fixed a priori) is slightly different: in that case, the choice of integration points reflects a modelling decision (i.e. the form of the movement kernel ), and has a direct impact on the interpretation of the parameters (as discussed by Beumer et al., 2023). This requires the additional assumption that movement is not affected by habitat selection, and the habitat selection parameters obtained from that method therefore only have the same interpretation as the output of other approaches if this assumption holds.…”