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

# Human access constrains optimal foraging and habitat availability in an avian generalist The dataset was collected in western Tennessee where GPS-marked mallards were monitored across three winters (November through March 2019-2022). We used conditional logistic regression to fit step selection functions to model resource selection by male and female mallards during winter across three periods (pre-season, hunting season, and post-hunting season) and night vs. day. We found that mallards selected areas free of human hunting and other disturbance (i.e., waterfowl sanctuary), which limited available habitat diurnally. However, mallards were freed from this constraint and selected "riskier" areas that were associated with hunting and other human disturbance nocturnally. Mallards selected sanctuary during pre-season, suggesting limited water availability and mallards did not immediately perceive human-accessible areas as safe during post-season, suggesting a lag-effect to reacclimate to non-hunting conditions. In general, mallards showed a functional response for high-energy (managed) habitat resources throughout seasons but especially nocturnally when the landscape was devoid of hunting and other human disturbance. ## Description of the data and file structure Due to size limitations, data is archived in a tabular format as an RData or ".rds" file. This file includes (1) mallard hourly telemetry locations for 426 individuals and nearly 43,000 "local-scale" flights and associated covariate vectors including (2) the surface water inundation; Allen 2016; (2) landcover layers for 2019, 2020, 2021 from publicly sourced datasets; (3) the managed habitat layers for each year that were field-sampled and computer digitized; and (4) the "human access" categorical variable depicting a gradient of hunting and other anthropogenic disturbance. All covariates were aligned to the same extent and resolution by resampling and stacking each .tif file and related to the locational data by extracting covariate values at each GPS location using the raster and terra package in R. To read the file into R, users need only to specify the path of the saved .rds file and use the function readRDS from base R. ## Sharing/Access information The raw GPS data is stored in the Movebank data repository and can be shared upon reasonable request. Geospatial data was derived from: The PAD-US public repository and U.S. Department of Agriculture's "Cropscape" Cropland Data Layer. Cropscape includes cropland land cover classes and non-agricultural land cover classes; the latter are derived from the National Landcover Database (Dewitz et al. 2021). These datasets are publicly available. ## Code/Software No new software or specialized functions were created. The authors used freely available R statistical software and packages including raster, terra to align .tif covariates and extract values to individual GPS locations. The authors used the clogit and amt packages to fit candidate models with robust standard errors (Therneau 2021, Signer et al. 2021). Last, the authors used the predict function in R to generate bootstrapped predictions for focal variables of interest, while holding non-focal variables at their mean values.