Abstract
Nutrition is a key driver of ungulate population dynamics and directly influences growth, reproduction, and survival. Given its importance, accurately quantifying the nutritional resources available to ungulates is essential for effective wildlife management. Traditional methods have often involved intensive field efforts to estimate nutritional metrics such as biomass, digestible energy, and digestible protein. Although foodscape models based on such data have been well supported, the requisite sampling is time and labor intensive. As a result, the Idaho Department of Fish and Game (IDFG) developed a fine-scale vegetation model (FSVM) that predicts the presence of understory plant species using remotely sensed covariates. We aimed to explore relationships between predictions of plant community composition derived from the FSVM and measured values of total (i.e., all understory vegetation) and high-quality (i.e., plant species and/or parts, such as leaves and flowers, that are high in energy or protein content) forage biomass. Using biomass measurements collected by the Clearwater Basin Collaborative in the St. Joe and Clearwater River basins during the summers of 2016 and 2017, we developed two generalized additive models (GAMs) using each forage metric in turn as a response variable. Our optimized base model for predicting high-quality forage biomass using a suite of remotely sensed covariates had an adjusted R² of 0.23, and our base model for predicting total forage biomass had an adjusted R² of 0.39. We then tested whether the addition of FSVM-derived covariates that were specific to deer and elk diets (i.e., plant community metrics that were tied to diet composition data for each species) would improve the base models. We found that FSVM covariates improved the predictive strength (adjusted R2) of foodscape models by a maximum of 2% across nutritional metrics (i.e., total and high-quality biomass), herbivore species (deer and elk), and spatial scales. Future research should explore whether the use of different FSVM covariates or alternative foodscape metrics (e.g., suitable forage biomass) could better capture fundamental relationships between plant community composition and nutritional resources. Our second goal was to evaluate whether and to what degree FSVM and/or nutritional covariates could be used to predict variation in relative abundances of elk and white-tailed deer in northern Idaho. Using images from 750 trail cameras deployed in IDFG’s game management units (GMUs) 1, 6, and 10A from 2020–2022 and a suite of forage covariates derived from the FSVM and our base GAMs, we modeled variation in the relative abundances of elk and white-tailed deer using Royle-Nichols abundance models fit within a Bayesian framework. Forage covariates were not useful for predicting variation in local abundance of elk, with the most supported model being the null model. In contrast, for white-tailed deer the inclusion of forage covariates, specifically high-quality forage biomass, which was positively related to relative abundance, did improve our models. Future work should explore scale dependency in relationships between forage availability and relative abundance to identify the scale at which these relationships are strongest for different ungulate species.
Our third goal was to assess how white-tailed deer in northern Idaho navigate trade-offs between accessing high-quality forage and avoiding human disturbance. The rapid expansion of human activity has led to wildlife population declines and observed changes in animal behavior on a global scale. Animal behavior plays a fundamental role in maintaining ecological processes such as nutrient cycling, primary productivity, seed dispersal, pollination, and pathogen transfer. The presence of humans, even in the absence of a direct threat (e.g., hunting), can lead to strong fear responses in wild animals, potentially resulting in suboptimal use of foraging habitats. We studied responses of white-tailed deer to human disturbance in IDFG’s game management units 6 and 10A from 2019–2022. We included four different candidate covariates to represent human disturbance in generalized linear mixed models of habitat use by deer: (1) Distance to motorized roads (i.e., roads/trails open to vehicular traffic); (2) Distance to non-motorized roads (i.e., roads/trails closed to vehicular traffic); (3) Number of unique cellphone GPS pings (e.g., only one ping per device counted); and (4) the number of total cellphone GPS pings (e.g., all pings counted regardless if they were from the same device). We also included model-predicted values of high-quality forage biomass and total forage biomass generated from our base GAMs (described above) as covariates in selection models. We evaluated two-way interactions between covariates to determine whether human disturbance modified use of the foodscape by deer (n = 161 deer).
Female white-tailed deer selected habitats that were closer to roads during summer, and when in close proximity to roads, deer tended to select habitats that had lower values of high-quality forage biomass. This pattern may partly reflect limitations in our base high-quality biomass GAM, which had an adjusted R² of 0.23. In addition, roads in our study area may overlap with preferred deer habitat, as they are often built in areas of lower elevation that have less rugged terrain. It’s also possible that road density in our study area was high enough that deer were simply unable to avoid roads. Additional tradeoffs, such as risk of predation or competition with elk may also influence this pattern.
We also tested for differences in selection by male deer between summer and fall (when human disturbance ostensibly increases due to the onset of hunting season), and between female deer with versus without a fawn at heel during early summer. We did not observe any behavioral differences in patterns of selection by males between summer and fall, nor between females with and without a fawn at heel. It is possible that differences in selection were occurring, but that the scale at which we chose to assess these relationships (the landscape scale, 2nd-order selection) was too broad to detect them. Future work should explore whether such behavioral shifts might be occurring at a finer scale.
In contrast to our results for roads, we found that white-tailed deer strongly avoided cellphone GPS pings, indicating that deer avoid humans in areas where they are most active. We consistently found that the number of unique device pings resulted in better model fit than the total number of pings (i.e., not tied to a device ID), suggesting that the presence of more individuals on the landscape may generate higher levels of disturbance than a smaller number of people who occupy an area for longer periods. Overall, cellphone GPS data provided a more nuanced and dynamic measure of human activity that was likely more representative of how people are using the landscape, and how that use influences wildlife. Moving forward, cellphone GPS data could be a powerful tool for wildlife managers wanting to assess the impacts of human activity on wildlife populations.