Mammalian Camera Trap Data; Northwest Arkansas
Data files
Apr 04, 2023 version files 1.59 MB
-
DRYAD_Data_Johansson.csv
-
README.md
Aug 14, 2023 version files 14.14 MB
-
DRYAD_Data_Johansson_Updated2.1.csv
-
README.md
Abstract
The human footprint is rapidly expanding, and wildlife habitat is continuously being converted to human residential properties. Surviving wildlife that reside in developing areas are displaced to nearby undeveloped areas. However, some animals can co-exist with humans and acquire the necessary resources (food, water, shelter) within the human environment. This may be particularly true when development is low intensity, as in residential suburban yards. Yards are individually managed “greenspaces” that can provide a range of food (e.g., bird feeders, compost, gardens), water (bird baths and garden ponds), and shelter resources (e.g., brush-piles, outbuildings) and are surrounded by varying landscape cover. To evaluate which residential landscape and yard features influence the richness and diversity of mammalian herbivores and mesopredators; we deployed wildlife game cameras in 46 residential yards in summer 2021 and 96 yards in summer 2022. We found that mesopredator diversity had a negative relationship with fences and was positively influenced by the number of bird feeders present in a yard. Mesopredator richness increased with the amount of forest within 400m of the camera. Herbivore diversity and richness were positively correlated to the area of forest within 400m surrounding yard and by garden area within yards, respectively. Our results suggest that while landscape does play a role in the presence of wildlife in a residential area, homeowners also have agency over the richness and diversity of mammals occurring in their yards based on the features they create or maintain on their properties.
Methods
2.1 Study Sites
Our study took place from 4 April to 4 August 2021 and 2022 within an 80.5 km radius of downtown Fayetteville, Arkansas USA. Northwest Arkansas is a rapidly developing area with a current population of approximately 349,000 people. Fayetteville is located in the Ozark Highlands ecoregion and the landscape is primarily forested by mixed hardwood trees with open areas used for cattle pastures and some scattered agriculture. Our study took place in residential yards ranging from downtown Fayetteville to yards situated in more rural areas. We solicited volunteers from the Arkansas Master Naturalist Program and the University of Arkansas Department of Biological Sciences who allowed us to place cameras in their yards. We attempted to choose yards that represented the continuum of urban to rural settings and provided a range of yard features to which wildlife was likely to respond to.
2.2 Camera Setup
To document the presence of wildlife in residential yards, we deployed motion-triggered wildlife cameras (Browning StrikeForce or Spypoint ForceDark) in numerous residential yards (46 yards in 2021 and 96 yards in 2022). We placed cameras approximately 0.95 m above the ground on either a tripod or a tree and at least 5 m from houses and at most 100 m from houses. When possible, we positioned cameras near features such as compost piles, water sources (natural or man-made), and fence lines to maximize detections of wildlife. We coordinated with homeowners to choose locations that would not interfere with yard maintenance or compromise homeowner privacy. We placed cameras in both front and back yards, although most cameras were placed in backyards. Backyards often had more features predicted to be of interest to wildlife (Belaire et al., 2015) and cameras placed in backyards were less likely to have false triggers associated with vehicular traffic or be vulnerable to theft. When necessary, we removed small amounts of vegetation that may have blocked the view of the camera or triggered the camera although this was always done on such a small scale as not alter the environment but just to clear the view of the camera. We set cameras to trigger with motion and take bursts of 3 photographs per trigger with a 5 s reset time. We did not use any bait or lures. We checked and downloaded cameras every 2 weeks to check batteries and download data. We moved cameras around the same yard upwards of 3 times within the season to ensure we captured the full range of wildlife present in each yard.
At each yard, we recorded eight variables associated with food, water, or shelter features in the yard area surrounding the camera, these variables were recorded in both front and backyard (Table 1). First, we recorded the area of maintained gardens occurring in each yard. Next, we recorded the volume of potential den sites available in each yard. Potential denning sites included the total available area under sheds and outbuildings as well as decking that was less than 0.3 meters off the ground and provided opportunities for wildlife to burrow beneath and be sheltered. Similarly, we also measured the volume of all brush and firewood piles present in each yard that could be used by smaller wildlife species for shelter or foraging. We counted the number of bird feeders in each yard that were regularly maintained during the study period. We also counted the number of water sources available including bird baths and garden ponds (any human subsidized water source on the ground usually within a lined basin or container). We distinguished between these types of water sources in analyses because bird baths were likely not available to all wildlife because of their height. We also categorized the presence and type of natural water source present in each yard including vernal streams, permanent streams or ponds, rivers, or lakes. We also recorded the presence of agricultural animals (such as chickens or ducks) and pets (type and indoor/outdoor) present in each yard (although we ultimately excluded the presence of pets from analyses – see below).
We documented whether the part of the yard where each camera was deployed was surrounded by a fence and if so, we categorized the fence type based upon its permeability to wildlife. We categorized fences into one of four categories ranging from those that posed little barrier to wildlife movement to those that were impassable to most species. For example, fences in our first category presented relatively little resistance to wildlife movement (i.e., barbed wire). A second category of fence consisted of fences made of semi-spaced wood slats or beams that offered enough room for most animals to squeeze through but that may have prevented passage of the largest bodied of the species. Fences that were about at least 1 m in height, but were closed off on the bottom (i.e., privacy or chain-link), meaning that few wildlife would be able to pass through without climbing or jumping over were placed in a third category. Finally, the fourth category of fences were those that were 1.8 m or greater in height and were made from a solid material that would prevent all wildlife except capable climbers from entering.
2.3 Landscape Variables
We used a GIS (ArcGIS Pro 10.2; ESRI, Inc. Redlands Inc) to plot the location of all cameras and to quantify the composition of the surrounding landscape. We first created 400m buffers around each camera, to encompass the average home range area of most wildlife species likely to occur in suburban yards (e.g., Trent and Rongtad, 1974; Hoffman and Gotschang, 1977; Atkins and Stott, 1998). Within each buffer, we calculated the amount of forest cover, developed open land (e.g., cemeteries, parks, and grass lawns), agriculture, and development using the 2016 National Land Cover Database (Dewitz 2019). We also quantified the maximum housing unit density (HUD) around each camera using the SILVIS Housing Data Layer (Hammer et al., 2004). Finally, we calculated the straight-line distance from each camera to the nearest downtown city center (Fayetteville, Rogers, Bentonville, or Eureka Springs). Distance to downtown is an additional index of urbanization and human activity that has been correlated with animal behavior in this area (DeGregorio et al. 2021).
Table 1 Description of all variables predicted to affect diversity and richness of mammals in residential yards within 80km of downtown Fayetteville, Arkansas USA during the April- August of 2021 and 2022.
Landscape Variables |
Variable Statistics |
||
Range |
Average ( 1 SD) |
||
Forest Cover |
Area of forest cover within 400m buffer |
0-0.45 |
0.18 0.13 |
Open Land |
Area of open land, (parks, cemeteries, and lawns) within 400m buffer |
0.003-0.31 |
0.09 0.06 |
Agricultural Land |
Area of land used for agricultural purposes within 400m buffer |
0-0.43 |
0.08 0.11 |
Developed Land |
Area of developed land within 400m buffer |
0-0.47 |
0.13 |
Housing Unit Density (HUD) |
Maximum Housing Unit Density within 400m buffer of camera (houses/ ) |
1-5095 |
657 |
Yard Variables |
|
|
|
Volume of Denning Sites |
Volume under sheds/outbuildings and under decks less than 1m off the ground ( ) |
0-700 |
27.3 |
Volume of Brush/Firewood Piles |
Total volume of denning sites including brush and firewood piles ( ) |
0-335.94 |
42.99 69.16 |
Water Source |
Number of human-maintained water sources |
0-7 |
1 |
- Bird Bath |
Water source that is raised off the ground, so much so that animals that cannot climb or jump cannot access it |
0-7 |
|
- Garden Pond |
Water source on or embedded within the ground |
0-3 |
0 1 |
Bird Feeder |
Number of bird feeders present in yard |
0-19 |
4 |
Garden |
Area of total maintained gardens ( ) |
0-525 |
46.13 85.13 |
Compost Pile |
Area of compost pile |
0-12 |
0.64 |
Fence Type |
If a camera was within a fence, it was given a score between 1-4, 1 being the most permeable fence and 4 being the most impassable to terrestrial mammals. 0: not in a fence 1: Barbed wire 2: Open slat fence 3: 1.2 m Chain-link or Privacy 4: 1.8 m chain-link or Privacy |
NA |
NA |
Poultry Presence |
Presence or absence of poultry being kept in yard |
NA |
NA |
Water |
Score of presence or absence of natural water source. 0: No natural water source 1: Vernal stream 2: Stream or pond 3: River 4: Lake |
NA |
NA |
Pets |
|
|
|
- Dogs |
Score of presence or absence of dogs: 0: No dogs 1: Indoor or leash walked 2: In fence or free roaming |
|
|
- Cats |
Score of presence or absence of cats: 0: No cats 1: Indoor only 2: Outdoor; at least partially |
|
|
2.4 Photo Processing
We used timelapse 2.0 (Greenberg et al. 2019) to sort and classify all wildlife photographs. We grouped photographs within 5 minutes to be counted as one sequence to reduce double counting individuals (Forrester et al., 2016). We extracted metadata (e.g., date, time) from photographs, determined the species present, and the number of individuals present in each sequence of photographs.
For our analyses, we focused on two guilds of mammals that are frequently encountered in yards and are reliably detected by cameras: mesopredators (medium-sized mammalian predators including raccoons, opossums, striped skunks, coyote (Canis latrans), bobcat (Lynx rufus), gray fox, red fox, and black bears (Ursus americanus)) and herbivores (white-tailed deer, cottontails, and woodchucks). This approach allowed us to assess how landscape and yard features affected a group of species that we anticipated used resources in similar ways. At each camera, we calculated the Simpson’s diversity (Simpson 1949) and richness of mesopredators and herbivores. Richness was defined as the number of species that were detected in a yard.
2.6 Statistical Analyses:
Before we began analyses, we conducted a collinearity test to evaluate relationships between variables. We considered two variables that had correlation coefficients |0.6| collinear. From those, we then decided which of the two variables were predicted to be more meaningful and only include that variable in subsequent analyses. We found that developed land and forest were highly correlated, r2= -0.706. Because we had a second measure of human development, housing unit density (HUD), already included we chose to keep forest cover going forward, the correlation between forest and HUD was r2= -0.46. We also found a high correlation, r2= 0.72 between the area of gardens and the volume of brush/firewood piles and subsequently removed brush/firewood piles from analyses. We also removed the pet variable from analyses because we felt that it did not capture the intended effect of cats and dogs on wildlife because the majority of yards (>80%) were regularly visited by cats and dogs even if the homeowner did not own cats or dogs. All other variables were retained for analyses. We scaled and centered all landscape variables on their mean to facilitate comparison between variables measured on different scales (Schielzeth 2010).
Because this study spanned two sampling years, we sampled forty-three individual yards in both years. To account for this repeated sampling, we adopted a conservative approach and randomly selected one year of monitoring for inclusion in analyses and excluded the other year.
To evaluate which landscape and yard variables most correlated with the Simpson’s diversity and richness of mesopredator and herbivore guilds recorded in yards, we used a Generalized Linear Model (GLM) analysis. We used both richness and diversity as response variables because they measure slightly different aspects of the wildlife community. Richness provides a coarse count of the number of species detected in a yard, while diversity provides a weighted measure of species in a yard accounting for both evenness and richness. We conducted four GLM analyses to explore the effects of landscape and yard features on the response variables of mesopredator diversity, mesopredator richness, herbivore diversity, and herbivore richness. For each analysis we used an iterative approach to assemble ninety-two candidate models (Table 5-8). The candidate model set for each analysis consisted of simple one-way variable models and all additive two-way combinations of the eight yard and four landscape predictor variables as well as a global model (including all additive variables) and a null model (Supplemental appendices 1-4). Using all two-way combinations allowed us to explore the effects of each variable while also assessing additive effects that could have been important for wildlife but without overparameterizing models.
For each analysis, we ranked candidate models using an information theoretic approach with Akaike’s Information Criterion corrected for small sample sizes (AICc). When appropriate, we derived parameter estimates for candidate models by model averaging all models within 3 ∆AICc (Burhnham and Anderson 2002) in R (R Core Team 2022) with the AICcmodavg package (Mazerolle MJ (2020).
To improve clarity in presenting model selection tables, we only display models that were competitive within 3 ΔAICc for each analysis (Table 2-4). Initial exploratory analyses indicated that relationships between predictor variables and response variables were linear and thus models were not corrected. Model goodness-of-fit was assessed using residual plots.