Human access constrains optimal foraging and habitat availability in an avian generalist

Animals balance costs of anti-predator behaviors with resource acquisition to minimize hunting and other mortality risks and maximize their physiological condition. This inherent trade-off between forage abundance and quality, and mortality risk is intensified in human-dominated landscapes because fragmentation, habitat loss, and degradation of natural vegetation communities is often coupled with artificially-enhanced vegetation (i.e., food plots) creating high-risk high-reward resource selection decisions. Our goal was to evaluate autumn–winter resource selection trade-offs for an intensively hunted avian generalist. We hypothesized human access was a reliable cue for hunting predation risk and thus predicted resource selection patterns would be spatiotemporally dependent upon levels of access and their perceived risk. Specifically, we evaluated resource selection of local-scale flights between diel periods of 426 mallards (Anas platyrhynchos) relative to wetland type, forage quality, and differing levels of human access across hunting and non-hunting seasons. Mallards selected areas that prohibited human access and generally avoided areas that allowed access diurnally, especially during hunting season. Mallards compensated by selecting for high-energy and greater quality foraging patches on allowable human access areas nocturnally when they were devoid of hunters. Post-season selection across human access gradients did not return to pre-hunting levels immediately, perhaps suggesting a delayed response to reacclimate to non-hunted activities and thus agreeing with the assessment mismatch hypothesis. Last, wetland availability and human access constrained selection for optimal natural forage quality (i.e., seed biomass and forage productivity) diurnally during pre-season and hunting season, respectively; however, mallards were freed from these constraints nocturnally during hunting season and during post-season. Our results suggest risk-avoidance of human accessible (i.e., hunted) areas is a primary driver of resource selection behaviors by mallards and could be a local to landscape-level process influencing distributions, instead of forage abundance and quality, which has long-been assumed by waterfowl conservation planners in North America. Broadly, even an avian generalist, well-adapted to anthropogenic landscapes, avoids areas where hunting and human access is allowed. Future conservation planning and implementation must consider management for recreational access (i.e., people) equally important as foraging habitat management for wintering waterfowl.

Methods
We captured male and female mallards in Tennessee from October–February 2019 through 2022. We banded ducks with U.S. Geological Survey aluminum tarsal bands and determined sex and age based on cloacal inversion, wing plumage and bill color (Carney 1992). We attached 20 g solar rechargeable and remotely programmable, OrniTrack Global Positioning System-Global System for Mobile transmitters (GPS-GSM; Ornitela, UAB Švitrigailos, Vilnius, Lithuania) to birds weighing ≥1,000 g to ensure deployment packages remained below 3 of an individuals’ body weight (Frair et al. 2010). We programmed GPS-GSM transmitters to record hourly locations throughout the duration of the study.

We filtered “used” locations to only those that were recorded within our spatial wetland extent data layer. Next, we generated 20-km circular buffers around used locations which corresponded to the maximum distance associated with local-scale flights (Appendix S1:Figure A1). We intersected 20-km buffers by our wetland data layer to ensure random locations were generated on available foraging habitat (i.e., on water) at each step. We then simulated 19 random locations within each buffer so each strata comprised 1 used location and 19 available locations for a 5% to 95% used to available ratio (de la Torre 2022).

We modeled resource selection using conditional logit models (i.e., discrete choice; Beatty et al. 2014a,b, Palumbo et al. 2019). We fit conditional logit models in the survival package using function clogit (Therneau 2020). We fitted separate candidate models for each season (i.e., pre-, hunting, and post-season) and diel period combination (i.e., diurnal and nocturnal) to account for and interpret variation in food depletion (Hagy and Kaminski 2015, Highway 2022), life-history events (e.g., pairing chronology; Heitmeyer 1985:268–269), and hunting mortality exposure (Palumbo et al. 2019; 6 candidate model sets × 2 diel periods × 3 seasons).