Why did the animal turn? Time-varying step selection analysis for inference between observed turning points in high frequency data

Abstract

Step selection analysis (SSA) is a fundamental technique for uncovering the drivers of animal movement decisions. Its typical use has been to view an animal as ''selecting'' each measured location, given its current (and possibly previous) locations. Although an animal is unlikely to make decisions precisely at the times its locations are measured, if data are gathered at a relatively low frequency (every few minutes or hours) this is often the best that can be done. Nowadays, though, tracking data is increasingly gathered at very high frequencies, e.g. >1Hz, so it may be possible to exploit these data to perform more behaviourally-meaningful step selection analysis. Here, we present a technique to do this. We first use an existing algorithm to determine the turning-points in an animal's movement path. We define a ''step'' to be a straight-line movement between successive turning-points. We then construct a generalised version of integrated SSA (iSSA), called time-varying iSSA (tiSSA), which deals with the fact that turning-points are usually irregularly spaced in time. We demonstrate the efficacy of tiSSA by application to data on free-ranging goats (Capra aegagrus hircus), comparing our results to those of regular iSSA with locations that are separated by a constant time-interval. Using (regular) iSSA with constant time-steps can give results that are misleading compared to using tiSSA with the actual turns made by the animals. Furthermore, tiSSA can be used to infer covariates that are dependent on the step-time, which is not possible with regular iSSA. As an example, we show that our study animals tend to spend less time between successive turns when the ground is rockier and/or the temperature is hotter. By constructing a step selection technique that works between observed turning-points of animals, we enable step selection to be used on high-frequency movement data, which are becoming increasingly prevalent in modern biologging studies. Furthermore, since turning-points can be viewed as decisions, our method places step selection analysis on a more behaviourally-meaningful footing compared to previous techniques.

A ‘How-to’ Guide for Interpreting Parameters in Habitat-Selection Analyses

“…This notation reflects the fact that step-selection parameters are scale dependent (i.e., different Δ t ’s will result in different estimates of β and γ ; see [14] for more details). Thus, we generally require observations to be equally spaced in time, and care must be taken when comparing inference from models fitted at different temporal resolution (but see [57]).…”

A ‘How to’ guide for interpreting parameters in habitat‐selection analyses

“…different Δt's will result in different estimates of and ; see Avgar et al, 2016 for more details). Thus,we generally require observations to be equally spaced in time (but seeMunden et al, 2020), and care must be taken when comparing inference from models fitted at different temporal resolution. When animals are observed at irregular time intervals, as with many marine species, it is possible to first fit a continuous-time movement model to the location data and then use this model to provide multiply imputed datasets that are regularly spaced in time (see e.g McClintock, 2017)…”

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

“…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 Ellison et al, 2024) and representations of previous space use (Schlägel & Lewis, 2014;Oliveira-Santos et al, 2016;Rheault et al, 2021).…”