Study design

The scientific goals of the NCCC project were to collect data to estimate white-tailed deer (Odocoileus virginianus) recruitment (i.e., fawn-doe ratios), coyote (Canis latrans) occupancy, and the distribution and abundance of multiple mammal species across the state. To meet these objectives, we aimed to get a representative sample of each county of North Carolina by sampling at least 28 locations per county (as suggested in Kays et al. 2020) every season (winter [December 1–February 28], spring [March 1–May 14], summer [May 15–July 31] and fall [August 1–November 30]), 2016 through 2019. For our idealized study design, we stratified sampling between public and private lands and within three main habitat types: open (a continuous area of at least 0.02 km² treeless land), forested (less than 0.02 km² treeless), and developed (impervious surfaces accounting for 80% to 100% of the total cover). Stratification goals within each habitat type were proportional to the habitat makeup of each county. For example, Carteret County is 61% forest, 11% developed, and 29% open, so we aimed to have at least 17 cameras set in forests, three in developed areas, and eight in open areas. While proportionality was a useful guideline during the project, final analyses can account for habitat effects as long as an adequate sample (40–60) per category is acquired (Kays et al. 2020).

Although we had our a priori study design, we did not tell volunteers where to run cameras specifically, but asked volunteers to choose a pre-selected project site or provide us with information on a site of their choosing. Though we did ask volunteers to avoid pointing cameras at roads and trails, we allowed them to choose a site on private property (92.2% of all NC lands are privately owned; Vincent et al. 2020) or from public land (7.8%; Vincent et al. 2020) where we already had secured permits. We worked with the National Park Service, the US Fish and Wildlife Service, the US Forest Service, North Carolina’s Wildlife Resources Commission (NCWRC), The Nature Conservancy(TNC), NC State Parks, and multiple smaller agencies to obtain 39 research permits that granted volunteers to access 175 separate public lands (with spatial scales anywhere from a few to several hundred acres) around the state at which volunteers were able to set cameras. Land managers sometimes had concerns about volunteers accessing permitted sites when the area might be scheduled for other events (e.g., logging or controlled burns); therefore, volunteers were required to contact managers before deploying cameras, giving managers a chance to deny access as needed. We also worked with 25 private landowners to allow volunteers access to large tracts of private land (i.e., volunteers had the option of setting cameras on these 25 private properties instead of on their own yard or public lands). Volunteers could choose from all of these potential camera locations (shared private or public lands) using an interactive map on the NCCC website. The map allowed users to zoom in on locations and search for a physical address or coordinates, and when a site was selected, it produced pop-up windows that listed information about the public land on which the site was located.

For cameras set at a site of volunteers’ choosing (often their yards), we asked volunteers to fill out a site description form that described aspects of their camera location that might affect local wildlife (hunting permissions, domestic animal presence, known animal feeding locations nearby, etc.). Each time they set a camera in a new location (deployment protocol described below), public or private, we asked volunteers to record deployment information (dates, location, and equipment type) and use the following field protocol:

1. Attach the camera to a tree at knee height, aimed parallel to the ground, at a relatively open area. Clear vegetation within 2m if necessary (to prevent false camera triggering).
2. Do not use bait or aim camera at features such as bird feeders, roads, game trails, or houses.
3. Camera settings should include multiple (3–5) pictures for each trigger with no delay between triggers.
4. If on a slope, the camera should face across the slope whenever possible.
5. Conduct a “walk test” to record the maximum distance the camera trap will trigger on a human (walk test protocol was included in training)
6. Obtain GPS coordinates of the camera’s exact location using a volunteer-supplied GPS device (i.e., smart phone).

We used un-baited motion- and heat-sensitive infrared flash trail cameras (mostly Reconyx HC 500 camera traps) to conduct the wildlife survey year-round for three years. Citizen scientists set camera traps for two- (fall) or three- (all other seasons) week deployments and spaced cameras at least 200m away from current or previous camera sites to maximize spatial coverage and reduce spatial correlation in counts (Kays et al. 2011). Volunteers could borrow one of our camera traps through North Carolina Public Libraries (Figure 1) or use their own, pre-approved camera trap model (camera traps with infrared flash and trigger speed less than 0.5 seconds were allowed). Volunteers who did use their own camera trap (33% of volunteers) used 28 different models from 10 camera trap manufacturers. 83.4% of deployments were run with Reconyx cameras, 9% run with Bushnell cameras, and the remaining run with cameras from 8 other manufacturers.

After obtaining a camera, volunteers chose a deployment site from our interactive map or deployed their camera on their private land and submitted a site description form. Once the two- or three-week deployment was complete, volunteers retrieved the camera, uploaded photos through the eMammal desktop application (McShea et al. 2016), and identified animals seen in the pictures (McShea et al. 2016; Figure 2a). We verified species identifications using eMammal’s online expert review application, after which the data were archived in the Smithsonian Digital Repository (Figure 2b). The majority of camera trap photos (excluding photos that possessed sensitive information–e.g., photos of endangered species or photos collected on certain private properties) and all photo metadata were made public immediately through the eMammal website (emammal.si.edu). To protect privacy, photos of people were never made available for public viewing. After photos were uploaded, volunteers could run their camera again at a different site or return the camera to the library from which they borrowed it. Three three-week camera trap deployment sessions were conducted in winter, spring, and summer. Though cameras could be borrowed at any time and set at will in between these deployment sessions, volunteers were asked to follow the deployment sessions as closely as possible to line up survey initiatives. We provided one-month breaks between seasons to give volunteers time to return cameras, if desired, or for new volunteers to go through training and prepare for the upcoming season.

During the fall, we focused sampling on ten representative counties for a deer-specific population study known as Fall Fawn Frenzy (FFF), which focused on estimating deer recruitment through fawn-doe ratios that can be used to help manage deer populations in NCWRC units. Focusing our cameras into ten representative counties concentrated our sample size at a time when fawns were most likely to be detected (i.e., fawns were more active as opposed to being mostly bedded, while spots on the fawns’ coats were still visible for easier fawn identification) so we could get more precise estimates (R. Kays, M. Lasky, B. Pease, A.W. Parsons, and K. Pacifici, in review). The fall season was also separated from the other seasons by a one-month hiatus, but deployments during the fall were two weeks in length (rather than the usual three weeks), and their timing differed by county. Thus, the sampling timeline can be summed up as three three-week deployment sessions in the winter, spring, and summer, and three two-week deployment sessions in the fall.

In addition to research goals outlined above, NCCC was also motivated by outreach and engagement objectives. The project provided a unique mechanism to connect a diverse network of volunteers around the state to wildlife science and conservation (Forrester et al. 2017), potentially changing the way the public thinks and acts with respect to wildlife. Achievement of this goal required creative and comprehensive approaches to volunteer recruitment and management, described in the section, “Volunteer recruitment and management”.

Volunteer recruitment and management

We hired one full-time volunteer coordinator and two part-time project managers to coordinate NCCC. Two to four part-time (5–10 hours per week, year-round) interns were also recruited to help ship cameras, upload data, and coordinate volunteers for the entire length of the project. Though anyone could participate in the project, we targeted recruitment efforts toward college students, naturalists, hikers, hunters, library patrons, and primary school teachers to obtain a wide variety of volunteers. Initial volunteer recruitment was achieved through press releases in local news sources across the state, a targeted Facebook advertisement, and an email campaign to NCWRC’s subscribers to eNews Blast. To continue to recruit volunteers throughout the project and to share results with existing volunteers, we sent out quarterly email newsletters, posted on social media outlets daily, and posted videos on YouTube

(both semi-annual webinars and various educational videos). We directly involved primary school classes in the project and created lesson plans that would allow NCCC to be implemented in secondary school (5^th–9^th grade) classes as part of the NC STEM curriculum (Schuttler et al. 2018). We managed volunteer information internally through the NCCC website, a series of linked Google Forms and Sheets, and later with a custom web application.

Our recruitment campaigns highlighted the project’s goals of learning about wildlife, especially in participants’ yards and neighborhoods. During the fall, we centered recruitment initiatives around white-tailed deer because of our FFF sub-survey. Recruitment techniques almost always included wildlife photos, and posed questions to the public such as “What wildlife is in your yard?”.

Participants signed up on the NCCC website and were then assigned a training module consisting of written instructions, video tutorials, and an online quiz. This training module took approximately 40 minutes to complete. Once trained, participants were asked to create an account on eMammal, an online data management system for camera trap data. eMammal offers a desktop application that allows volunteers to identify species and upload photos to a database that can later be accessed by researchers for quality control and, eventually, by the public to see the results of a project (McShea et al. 2016). After successful training, volunteers were approved to borrow a camera from a nearby library or use their personal camera if it possessed a trigger speed greater than 0.5 seconds and an infrared flash. We partnered with 63 libraries across North Carolina to distribute camera traps and accessories (lock, key, batteries, memory card) to volunteers (Figure 1). We used shared Google Sheets for librarians to log camera inventory information and to allow project staff to update the names of approved volunteers (i.e., those that had passed training) on an ongoing basis.

We sent prizes to volunteers after they achieved certain accomplishments to further encourage long-term participation (Jennett et al. 2016; Sarin and Mahajan 2001). These prize incentives were provided to volunteers at the beginning of the project, and volunteers were reminded of their existence in quarterly webinar updates. Drink snap-koozies were sent after the first deployment and a project-themed t-shirt was sent after participation for more than one year. To keep volunteers engaged, to share results, and to provide real-time feedback, we directed volunteers to eMammal’s auto-generated graphs and data overviews (see for an example data overview on the NCCC eMammal page, https://emammal.si.edu/north-carolinas-candid-critters). At the end of the project, we also created an interactive Tableau visualization on our website (Figure 3) to provide an overview of project findings.

To understand and enhance volunteer recruitment and retention (West and Pateman 2016), we integrated assessment of participation outcomes for volunteers engaged in NCCC into the project participation process. Everyone who successfully completed training was asked to complete this web survey prior to camera deployment. We then asked volunteers who deployed at least one camera to complete an optional post-project survey 6–12 months after their initial training completion date. Whereas pre-project surveys were distributed to align with the rolling project enrollment process, post-project surveys were distributed to certain cohorts of volunteers at pre-specified times (e.g., end of monitoring seasons) to facilitate administration and minimize time burden on project and research staff.

Study design

The NCCC study design made it easy for volunteers to run cameras at their preferred locations, but did not allow us to meet all a priori sampling goals. We found it difficult to obtain data on public lands, in rural areas, and with an even spatial coverage. Volunteers were more likely to set cameras on their own private property (54% of cameras set by volunteers were set on private lands), resulting in slightly fewer data points from public lands. We presume that the bias towards sampling on private land was a combination of volunteer curiosity regarding what animals lived on their property, the convenience of running cameras near home, and the increased difficulty surveying public lands owing to the need to coordinate with land managers. In addition, people sampling near their home were able to provide us with the location of the camera using an online map or from smart phone–obtained coordinates, rather than a handheld GPS unit, requiring less training and equipment. We assume the small sample size from rural areas (see relative distribution of camera traps in Figure 4a) is because fewer people live in these areas, and those living in urban areas are uninclined to travel to rural areas.

To address habitat and spatial sampling gaps (i.e., obtaining data on public lands, in rural areas, and with an even spatial coverage), we adopted a hybrid sampling design by monitoring where volunteers were setting cameras and supplementing data collection ourselves. To meet the objectives of FFF, we recruited the help of 3–5 NCWRC staff members in ten counties (30–50 individuals in total) and a lead NCWRC coordinator to set cameras across an even distribution of habitat types during the fall. With this assistance, 253 more cameras were set on public lands (7.4% of total deployments; Table 1). NCWRC staff assistance also led to a more even spatial coverage (see Table 1), and to acquiring 800,000 more photos (2.2 million total obtained in project) and 42,842 more wildlife observations (120,671 total) for NCCC (Table 1). Even with this assistance, the majority of all camera deployments remained set by citizen science volunteers (65% of all cameras were set by citizen scientists in all seasons, and 82% of cameras were set by citizen scientists outside of the fall season; Figure 4c). This combined effort resulted in a large sample distribution across the state (Figure 4a).

Despite staff filling in sampling gaps, our goal of proportional habitat sampling was not met in open habitats (18% sampled versus 26% intended; Figure 4b). We also did not achieve our seasonal goal of obtaining 28 deployments per county each season, which may partly be due to the concentration of deployments in ten specific counties during FFF. Nevertheless, we did–with our combined (i.e., mostly citizen scientist with supplemental professional camera trapping) sampling technique–achieve a minimum sample size across all habitat types of at least 200 deployments (forest = 2,248, open = 250, developed = 225), adequate to address our wildlife science objectives (i.e., estimating species distributions and coyote occupancy; R. Kays, M. Lasky, B. Pease, A.W. Parsons, and K. Pacifici, in review). Furthermore, the sample was tested for robustness and representativeness of all habitat types across the state. This test found that the data was both robust when sub-sampled for the project’s originally proposed species distribution occupancy models and had a large enough sample size to be representative of habitat categories on a statewide level as well as for 98.4% of land at the ecoregion level (R. Kays, M. Lasky, B. Pease, A.W. Parsons, and K. Pacifici, in review).

Citizen science projects face a tradeoff between getting more samples with fewer restrictions or using a strict sampling design that leads to more challenges in recruiting volunteers (Shirk et al. 2012). This may be less of a problem for small-scale studies; for example, Kays et al. (2017) was able to recruit volunteers to run cameras at specific pre-determined sites, but this targeted approach is difficult when the number of volunteers and geographic area increases. NCCC’s hybrid study design provided us with a large-scale dataset, including data from private lands (which represent more than 85% of the state) that are otherwise difficult to access. However, this required staff to monitor progress toward sampling goals during the study and supplement efforts with additional field work by project staff or associates, especially to sample in open habitats.

Based on our experience, we recommend that large projects needing a large amount of data allow more flexible sampling protocols that encourage participation. Once volunteers are engaged with the project, they can be encouraged to set sensors in specific areas through special programs or incentives to obtain data in certain locations or habitat types. For example, using a paired deployment strategy in which volunteers are asked to set one camera at an open site for every two cameras set in forested sites. Alternatively, managers could target participation by local conservation groups or clubs that can commit to meeting stratification goals before the project is initiated. This could allow project managers to realistically determine if their goals are going to be met or if they need to focus on a different approach. For smaller citizen science projects, requiring volunteers to follow a stricter protocol may make recruitment more difficult but will allow for more targeted data collection to fulfill project goals.

Volunteer recruitment and management