Some Memories Never Fade: Inferring Multi-Scale Memory Effects on Habitat Selection of a Migratory Ungulate Using Step-Selection Functions

Abstract

Understanding how animals use information about their environment to make movement decisions underpins our ability to explain drivers of and predict animal movement. Memory is the cognitive process that allows species to store information about experienced landscapes, however, remains an understudied topic in movement ecology. By studying how species select for familiar locations, visited recently and in the past, we can gain insight to how they store and use local information in multiple memory types. In this study, we analyzed the movements of a migratory mule deer (Odocoileus hemionus) population in the Piceance Basin of Colorado, United States to investigate the influence of spatial experience over different time scales on seasonal range habitat selection. We inferred the influence of short and long-term memory from the contribution to habitat selection of previous space use within the same season and during the prior year, respectively. We fit step-selection functions to GPS collar data from 32 female deer and tested the predictive ability of covariates representing current environmental conditions and both metrics of previous space use on habitat selection, inferring the latter as the influence of memory within and between seasons (summer vs. winter). Across individuals, models incorporating covariates representing both recent and past experience and environmental covariates performed best. In the top model, locations that had been previously visited within the same season and locations from previous seasons were more strongly selected relative to environmental covariates, which we interpret as evidence for the strong influence of both short- and long-term memory in driving seasonal range habitat selection. Further, the influence of previous space uses was stronger in the summer relative to winter, which is when deer in this population demonstrated strongest philopatry to their range. Our results suggest that mule deer update their seasonal range cognitive map in real time and retain long-term information about seasonal ranges, which supports the existing theory that memory is a mechanism leading to emergent space-use patterns such as site fidelity. Lastly, these findings provide novel insight into how species store and use information over different time scales.

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

“…KDE UDs do not consider the autocorrelation present in the GPS data (Fleming et al, 2015), but they are computationally efficient, and are practical to use in a simulation model where the previous space use density must be updated at every simulated time-step. The temporal decay component is similar to the decaying parameter function of in Merkle et al (2014) and Rheault et al (2021), but the negative exponential decay parameter that we used initially declines slower and has a lighter tail. Either function is likely to be appropriate for most situations.…”

“…The denominator normalises over the temporal decay component. Similarly to Rheault et al (2021), we excluded locations from the past 24 hours, as these locations are likely to reflect the autocorrelation in the movement process rather than resulting from a memory process.…”

“…Including a covariate that allows for a tendency to return to previously visited locations, which represents a memory process, is becoming more common in step selection models and is known to provide less-biased inference on external selection parameters when there is a clear site-fidelity or home range behaviour (Schlägel & Lewis, 2014; Oliveira-Santos et al, 2016; Rheault et al, 2021). We incorporated location history through the density of previous locations, which is similar to the approaches of Oliveira-Santos et al (2016) and Rheault et al (2021). We used kernel density estimation (KDE) to estimate the utilisation density (UD) of previous locations, which we combined with a temporal decay component.…”

“…The denominator normalises over the temporal decay component. Similarly to Rheault et al (2021), we excluded locations from the past 24 hours, as these locations are likely to reflect the autocorrelation in the movement process rather than resulting from a memory process. We used the observed data to estimate the σ parameter with the normal or 'reference' bandwidth estimator over all used points using the amt package , and σ was estimated separately for each individual.…”

Simulating animal space use from fitted integrated Step‐Selection Functions (iSSF)

“…We expect the recent interest in simulations from integrated step selection functions to continue. Recent extensions to iSSAs that include memory (Rheault et al, 2021), behavioural states (Klappstein et al, 2022;Pohle et al, 2023) or irregular sampling rates (Kim et al, 2023;Munden et al, 2021) could eventually be incorporated into the simulator for even greater realism.…”