Black‐tailed deer seasonal habitat selection: accounting for missing global positioning system fixes

Abstract

Quantitative models of black-tailed deer (Odocoileus hemionus columbianus) habitat selection do not exist for Pacific Northwest landscapes but are needed to make predictions of how deer respond to timber harvest. We developed models to estimate habitat selection by black-tailed deer within seasonal home ranges in the west-central Cascade Mountains, Washington, USA, from 2008-2019. Even with moderately high global positioning system (GPS) fix success within summer and winter (both averages ≥0.87), we were concerned about landscape or vegetation characteristics causing fix success failure that would bias estimates of habitat selection. Year of GPS-collar production, deer activity, and diel period (night, day, crepuscular) were associated with fix success. The odds of acquiring a fix were 2.53 times greater when the deer were inactive compared to active, and this difference was greater for those deer with higher average percent canopy cover within their home range.These relationships only explained a portion of the variation in the data and did not negate concerns about the effects of missing locations when estimating habitat selection, so we estimated deer habitat selection within summer and winter using an integrated step selection function (iSSF) and a probability of detection integrated step selection function (PDiSSF). The iSSF ignored missing fixes; the PDiSSF accounted

Methods for Implementing Integrated Step-Selection Functions with Incomplete Data

“…Our study contributes to the growing body of literature that extends iSSFs and improves the method's robustness under various conditions. This includes approaches for modeling irregular data (Munden et al, 2021), accounting for spatial dependence among residuals (Arce Guillen et al, 2023), methodological frameworks for fitting iSSFs with random slopes (Muff et al, 2020), as well as incorporating the probability of successfully obtaining an animal location in different habitat conditions (Vales et al, 2022).…”

“…It is similar to, but slightly less principled, than the approach developed by Vales et al (2022), which formally constructs the likelihood for multistep-durations by integrating out the missing steps. An advantage of this latter approach is that one can also attempt to account for non-random missingness by explicitly modeling factors related to the probability of obtaining a successful location (Vales et al, 2022). Nonetheless, integrating over the missing steps, as in Vales et al (2022), can be computationally intensive and prohibitive with large data sets.…”

“…An advantage of this latter approach is that one can also attempt to account for non-random missingness by explicitly modeling factors related to the probability of obtaining a successful location (Vales et al, 2022). Nonetheless, integrating over the missing steps, as in Vales et al (2022), can be computationally intensive and prohibitive with large data sets. Another downside of both of these approaches (the multistep approach and the approach of Vales et al, 2022) is that they can only be applied in cases where step-durations are a fixed multiple of the regular step-duration; i.e., unlike the dynamic+model approach, they cannot be applied when data are highly irregular.…”

“…Methods commonly used to analyze animal movement data, including step-selection analyses (Fortin et al, 2005; Thurfjell et al, 2014; Fieberg et al, 2021) and hidden Markov models (Michelot et al, 2016), require animal locations (terms in bold at first occurrence are defined in Table 1) that are collected at a constant sampling frequency, leading to data that are equally spaced in time. Yet, it is common to encounter missing locations in most telemetry data sets (Frair et al, 2010; Hofman et al, 2019; Vales et al, 2022), which introduces unwanted irregularities in the duration between successive locations. Thus, there is a need for analytical tools that enable the analysis of such data, while mitigating potential biases arising from temporal irregularity introduced through missing animal locations.…”

Role and Development of an Android App for the Protection of Wildlife

“…This information may be used by the receiver to determine the animal's location as well as its direction of travel [7]. By employing the Global Positioning System (GPS), optional environmental sensors, or automated data retrieval, biologists, scientific researchers, or conservation organizations can examine relatively fine-scale movement or migratory patterns in a free-ranging wild animal [8]. Animal tracking data enables us to comprehend how individuals and populations travel within their immediate environment, migrate across oceans and continents, and change over the course of generations [9].…”